Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
SECURE COMMUNICATION USING
STEGANOGRAPHY IN IMAGE PROCESSING
MUHAMMAD ZAHEER
A Thesis
Submitted in Partial Fulfillment of the requirements for the Degree of
Doctor of Philosophy
DEPARTMENT OF ELECTRICAL ENGINEERING AIR UNIVERSITY
2017
SECURE COMMUNICATION USING
STEGANOGRAPHY IN IMAGE PROCESSING
Ph.D. Dissertation
SUBMITTED BY MUHAMMAD ZAHEER REG. No. Ph.D.-EE-110916
SUPERVISOR
PROF. DR. IJAZ MANSOOR QURESHI
DEPARTMENT OF ELECTRICAL ENGINEERING AIR UNIVERSITY
ISLAMABAD 2017
CERTIFICATE OF APPROVAL
Department of Electrical Engineering
It is hereby certified that Muhammad Zaheer (Reg # Ph.D.-EE-110916) has successfully completed
his dissertation.
_____________________________
Dr. Ijaz Mansoor Qureshi Air University
Supervisor
____________________________ ____________________________
Dr. Fida Muhammad Khan Dr. Waseem Khan
Internal Examiner 1 Internal Examiner 2
Guidance and Evaluation Committee Guidance and Evaluation Committee
____________________________
Dr.Adnan Omer
External Examiner
Guidance and Evaluation Committee
____________________________ ___________________________
AVM Saleem Tariq Dr. Zafar Ullah Koreshi
Chair Department Dean
ABSTRACT
Research in information security and secrecy is becoming more and more important, as well as,
demanding as the information is exponentially exploding. There has been a good bit of focus on
cryptography, but with cryptanalysis and crypto attacks, researches have looked into the
alternative means, like steganography. Steganography conceals the message into the cover file.
In this dissertation we have focused on image steganography and we have tried to improve the
key parameters of the system which are capacity, imperceptibility and robustness.
In the first part of the thesis the presented work comprises of two contributions. In the first
contribution the secret information is preprocessed by using latest devised right translated gray
substitution box and BCH error correction codes. This process enhances the security of the
information as data retrieval is impossible without the information of mapping rule applied and
secret key. Differential embedding to the LSBs of optimized chaotically selected pixels through
GA is another step to make detection difficult and avoid error to propagate. In second
contribution we introduce an advance technique for embedding based on representing cover
image pixels using a new 13-bit prime series representation resulting into 3-times increase in
capacity. The payload has been reduced by applying 2D DCT to secret image and thresholding of
the coefficients ensuring high imperceptibility. An innovative algorithm has also been proposed
to ensure the uniform spread of message into the cover image by adjusting average separation
between the chaotically selected pixels of cover image for embedding based on size of secret
information and cover image
In the second part of the thesis the presented work is an application of compressed sensing. The
main advantage we have gained is the huge increase in security of the information along with the
payload reduction. Utilizing compressed sensing made the system more secure as reconstruction
of the data is impossible without knowing the measurement basis, as the generation process of
measurement basis is random and requires huge computations to predict them. We have
presented two problems. The first contribution focuses on combining compressive sensing and
steganography to enhance security of image steganography. The secret image is encrypted and
compressed using compressive sensing. The encrypted data is then embedded to randomly
selected pixels of cover image using a secret chaotic key. Simulation results presented show the
efficient recovery and reconstruction of secret image using reduced payload. In the second
contribution we have focused on security and payload capacity enhancement of an image
steganography system for an audio message by using compressed sensing theory. However, in
order to utilize compressed sensing, the audio message is first converted to an equivalent
grayscale image which is sparsified using 2D-DCT and thresholding. The sparsified image is
further compressed using the proposed compressed sensing algorithm which enhances the
security to a high level and also payload capacity improves significantly, without losing
imperceptibility of the system. The compressed image is embedded in chaotically chosen pixels
of the cover image. At receiver the compressed sensing reconstruction algorithm is used to
reconstruct the grayscale image which is then converted back to the audio message. Presented
results indicate that the proposed system is highly imperceptible, secure and robust against
various image processing attacks. It is able to reconstruct secret audio message with high PSNR
value.
Copyright by Muhammad Zaheer 2017 All rights reserved. No part of material protected by this copyright notice may be reproduced or utilized in any form by any means, electronic or mechanical, including ohotocopying, recording or by any information storage and retrieval system, without permission from the author.
i
Publications and Submissions
• Muhammad Zaheer , I.M.Qureshi, Zeeshan Muzaffar and Laeeq Aslam, Compressed Sensing Based Image Steganography System for Secure Transmission of Audio Message with Enhanced Security”, International Journal of Computer Science and Network Security (IJCSNS) (ISI Indexed; Accepted and Published)
• Muhammad Zaheer, I.M.Qureshi, Zeeshan Muzaffar, Laeeq Aslam, “High Capacity Image Steganography Based on Prime Series Representation and Payload Redundancy Removal”, (KSII Transactions on Internet and Information System.(IF=0.33: Under Review)
• Muhammad Zaheer, I.M.Qureshi, M.Zeeshan Muzaffar, Tahir Naseem, “Compressed Sensing Based Improved Imperceptibility and Security in Image Steganography System ( Springer IJIS, IF= 1.2, Under Review)
• Muhammad Zaheer, I.M.Qureshi , Tahir Naseem and Zeeshan Muzaffar, “Improved and Secure Differential LSB Embedding Steganography Based on Chaos, Genetic Algorithm and BCH Codes( JIAS , ISI Indexed, Under Review)
ii
Acknowledgements First and above all, I would like to express my gratitude to Almighty Allah for providing me the blessing to complete this Thesis. I would like also to express my sincere appreciation to my supervisor Prof.Dr.Ijaz Mansoor Qureshi. His guidance and assistance throughout the research was the reason behind accomplishing this work. His advice and suggestions were always beneficial. Further I would like to thank Mr. Zeeshan Muzaffar, Mr. Laeeq Aslam, and Dr. Tahir Naseem for assisting me in different issues we faced during this research. I would like to express my deepest thankfulness to my parents. Their continued encouragements led me to the correct way. Special thanks to my brothers for their continuous support. I dedicate this master thesis to my wife and son who always have helped me and believed that I could do it.
iii
Table of Contents
Publications amd
Submissions…………………………………………………………………………………….....i
Acknowledgements…. …………………………………………………………………………..ii
List of Figures…………………………………………………………………………………...vi
List of Tables…………………………………………………………………………………...viii
List of Abbreviations……………………………………………………………………………ix
List of Symbols…………………………………………………………………………………..x
Chapter 1………………………………………………………………………………………..01
1.1: Introduction …………………………………………………………………………………01
1.2: Problem Formulation ……………………………………………………………………….03
1.3: Contributions ……………………………………………………...………………………..04
1.4: Thesis Outline ………………………………………………………………………………05
Chapter 2 ………………..……………………………………………………………………...07
2.1:Information Hiding ………………………………………………………………………….07
2.2:Steganography ………………………………………………………………………………10
2.2.1 :Text steganography ……………………………………………………………….10
2.2.2: Image steganography ……………………………………………………………..10
2.2.3: Audio steganography ……………………………………………………………..11
2.2.4: Video steganography ……………………………………………………………..11
2.2.5: Protocol steganography …………………………………………………………...11
2.3: Image Steganography……………………………………………………………………….11
2.3.1: Spatial Domain……………………………………………………………………12
2.3.2: Transform Domain ………………………………………………………………..12
2.4: Related Work………………………………………………………………………………..12
2.5:Summary ……………………………………………………………………………………20
Chapter 3 ………..……………………………………………………………………………...21
3.1: Introduction …………………………………………………………………………………21
3.2: Compressed Sensing ………………………………………………………………………..21
iv
3.2.1: Sparsity …………………………………………………………………………...22
3.2.2: Incoherent Sampling ……………………………………………………………...23
3.2.3: Restricted Isometry Property ……………………………………………………..24
3.3: Substitution-Boxes …………………………………………………………………………26
3.4: Chaotic Key Generation ……………………………………………………………………27
3.4.1: Chaotic Systems ………………………………………………………………….27
3.4.2:The Logistic Map System …………………………………………………………27
3.4.3: Utilizating Logistic Map in Image Steganography………………………………..28
Chapter 4 ………………………..……………………………………………………………...30
4.1: Introduction …………………………………………………………………………………30
4.2: Improved and Secure Differential LSB Embedding Steganography Based on Chaos, Genetic
Algorithm and BCH Codes ……………………………………………………………………..31
4.2.1: Proposed Model …………………………………………………………………..31
4.2.2: Simulation Results ………………………………………………………………..39
4.3: High Capacity Image Steganography Based on Prime Series Representation and Payload
Redundancy Removal …………………………………………………………………………...42
4.3.1: Proposed Model …………………………………………………………………..42
4.3.2: Proposed System Implementation ………………………………………………..44
4.3.3: Extraction and Recovery of Message …………………………………………….48
4.3.4: Simulation Results ………………………………………………………………..49
4.4: Summary ……………………………..……………………………………………………..54
Chapter 5 ………..……………………………………………………………………………...55
5.1: Introduction …………………………………………………………………………………55
5.2: Compressed Sensing Based Improved Imperceptibility and Security in Image Steganography
System …………………………………………………………………………………………...57
5.2.1:Proposed Model …………………………………………………………………...57
5.2.2: Advantages of Using Compressed Sensing in Image Steganography…………….60
5.2.2: Simulation Results and Discussion………………………………………………. 61
5.3: Compressed Sensing Based Image Steganography System for Secure Transmission of
Audio Message with Enhanced Security ………………………………………………………..67
5.3.1: Transmitter Side …………………………………………………………………..67
v
5.3.2: Decoding and Reconstruction of Secret Message ………………………………...69
5.3.3:Simulation Results ………………………………………………………………...71
5.4: Summary ……………………………………………………………………………81
Chapter 6 ………………………………..……………………………………………………...83
6.1: Conclusions …………………………………………………………………………………83
6.2: Future Work ………………………………………………………………………………...85
References……………………………………………………………………………………….86
vi
List of Figures
Figure 2.1: Categorization of Information Hidding Techniques …………………………………7
Figure 2.2: Types of Steganography …………………………………………………………….10
Figure 3.1: Sparsity in Images …………………………………………………………………..22
Figure 3.2: Compressed Sensing Model………………………………………………………....25
Figure 3.3: Recovery Algorithms………………………………………………………………..25
Figure 3.4: An Example of S-Box ………………………………………………………………27
Figure 4.1: Proposed Method ……………………………………………………………………32
Figure 4.2: Right Translated Gray S-Box ……………………………………………………….33
Figure 4.3:Optimized Chaotic Key Generation using GA ………………………………………37
Figure 4.4: Differential LSB Encoder …………………………………………………………...38
Figure 4.5: Proposed Decoder …………………………………………………………………...39
Figure 4.6: Original and Stego Image Flower …………………………………………………..40
Figure 4.7: Original and Stego Image Baboon ………………………………………………….41
Figure 4.8: Original and Stego Image Tree ……………………………………………………..41
Figure 4.9: Original and Stego Image Man ……………………………………………………..41
Figure 4.10: Block Diagram of Proposed Model ………………………………………………..43
Figure 4.11: Chaotic Key Generator …………………………………………………………….44
Figure 4.12: Adaptive Key Generation ………………………………………………………….45
Figure 4.13: Embedding System ………………………………………………………………...47
Figure 4.14: Extraction and Message Recovery ………………………………………………...48
Figure 4.15: (a) Message Image Eagle (b) Message Image Roger ……………………………...49
Figure 4.16: Cover Images (a) Baboon (b) Lena ………………………………………………..50
Figure 4.17: Recovered Message Image Eagle ………………………………………………….52
Figure 4.18: Recovered Message Image Roger …………………………………………………53
Figure 5.1: Encoder ……………………………………………………………………………...58
Figure 5.2: Decoder ……………………………………………………………………………..59
Figure 5.3: Original Cover and Secret Images ………………………………………………….62
Figure 5.4(a-e): Reconstructed Secret Image with m=40, 80, 120, 160 and 200 ……………….63
Figure 5.5(a-e): Reconstructed Secret Image with m=40, 80, 120, 160 and 200 ……………….65
vii
Figure 5.6: Secrecy Enhancing and Embedding of the Secret Message Using CS .......………...67
Figure 5.7: De-Embedding and Decoding of Image using CS Recovery Algorithm …………...69
Figure 5.8: Cover Image ………………………………………………………………………...71
Figure 5.9: Grayscale Image generated against audio message …………………………………71
Figure 5.10: PSNR of audio message recovered with Gaussian noise added …………………..73
Figure 5.11: Stego Image with Gaussian Noise added noise variance …………………………74
Figure 5.12: PSNR of audio message recovered with salt & pepper noise added ………………75
Figure 5.13: Stego Image with Salt & Pepper Noise added at noise variance …………………..77
Figure 5.14: PSNR of audio message with speckle noise ……………………………………….78
Figure 5.15: Stego Image with Speckle Noise added at noise variance ………………………...79
Figure 5.16: PSNR of recovered audio with Poison Noise added ………………………………80
Figure 5.17: (a) Stego Image with no noise (b) With poison noise ……………………………..81
viii
List of Tables
Table 2.1: Comparison of Information Hiding techniques ……………………………………….8
Table 4.1: PSNR of different cover images without using GA optimization …………………...42
Table 4.2: PSNR of different cover images using GA optimization ……………………………42
Table 4.3: Payload and Average Separation Calculation for Message Image Eagle ……………50
Table 4.4: Payload and Average Separation Calculation for Message Image Roger …………...50
Table 4.5: Simulation Results for Message Image Eagle ……………………………………….53
Table 4.6: Simulation Results for Message Image Roger ……………………………………….53
Table 5.1: PSNR of Reconstructed Handwritten Secret Image …………………………………64
Table 5.2: PSNR of Reconstructed Typed Secret Image ………………………………………..65
Table 5.3: Comparison of proposed scheme with scheme presented in Section 4.3 in terms of
payload reduction………………………………………………………………………………...66
Table 5.4: PSNR of recovered audio in dB with added Gaussian noise ………………………...72
Table 5.5: PSNR of stego Image in dB with Gaussian Noise …………………………………...73
Table 5.6: PSNR of audio message in dB with:Salt and Pepper noise ………………………….75
Table 5.7: PSNR of stego image in dB with Salt and Pepper noise ……………………………76
Table 5.8: PSNR of audio message in dB with Speckle noise …………………………………..77
Table 5.9: PSNR of Stego Image in dB with speckle noise ……………………………………..78
Table 5.10: PSNR of audio message in dB with Poison noise added …………………………...80
Table 5.11: PSNR of Stego Image in dB with Poison noise added……………………………...80
ix
LIST OF ABBREVIATIONS
PSNR Peak Signal to Noise Ratio SSIM Structural Similarity Index C Cover Image K Secret Key S Stego Image CS Compressed Sensing AES Advanced Encryption Standard DES Data Encrytion Standard BCH Bose–Chaudhuri–Hocquenghem MSE Mean Square Error Th Threshold
x
LIST OF SYMBOLS Φ Measurement Basis Ψ Representation Basis
m Number of rows in Φ � Orignal Image s Sparse Indicator
µ Coherence
ε Variance of measurement noise
2
Chapter 1
INTRODUCTION
1.1 INTRODUCTION
Nowadays, individuals exchange information using the existing communication technologies
such as a local area network or some communication is done via wide area network [1]. This
information can be very sensitive and needs to be protected against any intruder, who can
intercept it during the communication phase [2]. Therefore, transferring secure information
cannot be solely relied on the existing communication technologies. We need robust techniques
in order to protect the information and ensuring its secure transfer from one point to another [3],
[4].
Cryptography is used for the encryption of secret information based on some mathematical
relationships. Cryptography has been widely used for the protecting the information that is
exchanged via Internet. Mailing servers and website designing widely use public channels for
transferring of information using cryptography [5].
However, both technologies are vulnerable to intrusion and attacks and exchanged information
can be easily retrieved. In cryptography the secret information is modified to a new form using
some public or private keys and the modified information becomes unreadable (e.g., encryption).
This encrypted information is then sent over the public communication channels and at the
receiver side the original information can only be retrieved using the corresponding keys (e.g.,
decryption) [6]. This technique cannot prevent against attacks that can intercept the decrypted
information and apply some other techniques to retrieve the secret information. Therefore, it is
necessary to find alternative methodology to protect the information exchanged over public
channels without raising suspicions [7]. This methodology is considered to be steganography and
has become very popular in the last decade.
3
Steganography is an algorithm based on of concealing secret message into a digital cover (i.e.,
images, audio, and text). It completely differs from the cryptography algorithms [8], [9]. In fact,
in cryptography the secret information is modified but still can be seen in an unreadable format
once sent over the data networks, whereas steganography works on embedding information into
a digital cover and cannot be perceived as long as the imperceptibility of the cover is not
deteriorated [10]. The steganography technique has been utilized for many years to convey secret
messages. For instance, a king in ancient Greece had a practice of shaving the head of a slave
and the required message to be sent was written on it. When the hair was grown, the slave was
sent to the destination to distribute the secret message. The receiver then shaved the hair and got
the secret message [11].
In current communication world, the main applications of steganography include embedding
copyright, embedding individual’s detail in smart IDs and inserting patient detail in medical
imaging system. A rapid growth of interest in steganography based algorithms has been noticed
in the applications related to military institutions.
Steganography conceals information in a digital medium called cover medium which can include
a video clip, an image, any audio file or simply a text which are called a cover image, a cover
audio, a cover video, and a cover text, respectively. After embedding the secret information the
related cover is termed as a stego-object [12]. If the cover is an audio file or a digital image, then
after embedding the cover word is replaced by stego resulting in to stego-audio or stego-image,
respectively.
Literature shows that images are considered as carriers providing high performance for hiding
and transmitting secret information over networks. Currently many algorithms have been devised
in order to use images for hiding secret information without disturbing the quality [13]. In this
thesis we focus on image steganography algorithms. An image consists of light luminance or
pixels represented as an array of values at different points. A pixel usually consists of one byte or
more. For example in 8-bit format images each pixel is defined as a combination of 8-bits or one
byte, whereas each pixel when represented in a 24-bit image uses three byte for representation
Red, Green and Blue (RGB) colors. Any variation of the bits leads to a new shade of color.
4
In a good steganography algorithm, there are five vital features that should be considered. The
first one is the payload capacity, defined as the total or maximum secret information that a stego-
cover can carry without noticeable distortions [14], [15]. The second feature is the un-
detectability which ensures that embedding should not be detectable whenever the stego-object is
analyzed. Other features that are considered include: invisibility, security and robustness [16].
1.2 PROBLEM FORMULATION
Over the past few years, Steganographers have achieved a massive success in hiding confidential
data from the attackers and intruders. Currently available steganography algorithms mainly focus
on the strategy used for embedding but comparatively less importance has been given to the pre-
processing stage, including encryption, reduction in payload and capacity improvement as the
dependence of these algorithms is on the encryption standards which are not designed
specifically to be utilized in steganography, where some extra features like flexibility, robustness
and security need to be addressed. Keeping the research directions as a step forward, there exists
adequate scope for improvement in related research [17]. The previous work in steganography
was not able to achieve high payload capacity while simultaneously improving the security of the
system. In this thesis, the main focus is to propose different techniques in image steganography
which focus on improving the payload capacity and secrecy of the information at the same time.
We have tried to efficiently utilize combination of steganography, cryptography and compressed
sensing for ensuring high payload capacity, imperceptibility and robustness of the system.
In particular, the objectives are narrowed down as follows:
1. To achieve high payload capacity keeping in mind the need to embed more information
but maintaining a good imperceptibility in order to make the system less susceptible to
the intruders
2. Exploring modern and state of the art methods to be incorporated in the steganography to
make secret information more secure and robust against different attacks
3. Using methodologies that can simultaneously increase the payload capacity,
imperceptibility and security of the system
5
1.3 CONTRIBUTIONS
This work focuses on presenting some novel techniques for performance enhancement of image
steganography. The key focus is on the development and validation of a novel approach to
enhance the performance of steganography methods and compare them with the ones in the
literature.
The main contributions of this work are Chapter 4 and Chapter 5. The first two without
compressed sensing, are from Chapter 4. The latter two are with compressed sensing and are
from Chapter 5.
• Incorporation of state of the art encryption technique based on gray code translated S-
Boxes and Reed Solomon error correcting codes to the image steganography system for
improving the secrecy of the information before embedding. The secret text information
has been pre-processed by encrypting and then providing a cover of error correction
codes to improve secrecy and error free recovery. Differential LSB encoding has been
introduced for embedding information in chaotically selected pixels instead of
conventional LSB in order to remove error propagation and perform at the same level as
LSB embedding in terms of imperceptibility.
• A novel algorithm has been proposed to automatically calculate the average separation
required between the pixels of the cover image that are chaotically chosen to embed the
secret information. To enhance the payload capacity a new embedding method has been
proposed which utilizes a novel prime series representation of the cover image pixels.
The proposed embedding scheme provides three times capacity as compared to the
conventional LSB embedding and a better secrecy of the information.
• Compressed sensing (CS) has been integrated with image steganography to attain the
research objectives of this work efficiently. Compressed sensing (CS) has been applied to
the secret information and this processed information is embedded into the cover image.
Compressed sensing not only enhances the secrecy of the information which is the main
gain but also reduces the payload for same amount of information. The measurement
basis used in compressed sensing are randomly generated and only known to the receiver.
6
The dimensionality of the measurement basis and the possible combinations to generate
them makes it impossible for the attacker to predict them and hence it is impossible to
reconstruct the compressed information. The above mentioned features of CS make it a
very lucrative choice for steganography in order to improve the capacity and secrecy of
the steganography algorithm at the same time. We have proposed a compressed sensing
based image steganography system and we have shown that the compressed secret
information has been successfully reconstructed from the stego image at receiver side
with high PSNR. Due to reduced payload the PSNR of stego image is also ensuring high
imperceptibility.
• The above proposed framework based on CS for images has also been efficiently utilized
to secretly transmit an audio message. In order to use the proposed work for an audio
message the audio message has been first converted to a grayscale image and then the
proposed compressed sensing framework for images has been applied on this grayscale
image. The presented results show that the audio message is also successfully recovered
with high PSNR even after applying different image processing attacks.
1.4 THESIS OUTLINE
The rest of the thesis has been organized as follows:
Chapter 2 starts with general introduction of the information hiding techniques, compares
different techniques and then focuses on steganography. Image steganography and the related
work in this field has been discussed in detail.
Chapter 3 provides an overview of the mathematical techniques that are being used in this work.
Compressed sensing, S-Boxes and chaotic key generation has been discussed in order to give an
overview of these techniques which are used in the later chapters.
Chapter 4 presents two contributions of the thesis in detail. Firstly the technique based on gray
translated S-Boxes, differential LSB embedding and RS error correcting codes has been
discussed and then we have presented the technique based on novel 13-bit prime series
representation system to improve the payload capacity significantly
7
Chapter 5 is based on the details of the contributions of this work that are based on compressive
sensing. CS based image hiding technique is presented first indicating performance enhancement
considering metrics like secrecy, payload capacity and imperceptibility. CS based audio hiding
technique is presented later in this chapter, which is based on the technique used for images and
it has been shown that the technique is able to perfectly reconstruct the audio message as well,
while ensuring high capacity, secrecy and imperceptibility.
Chapter 6 Summarizes the thesis and presents the future directions.
8
Chapter 2
LITERATURE REVIEW
2.1 INFORMATION HIDING
With the swift growth of information technology, information hiding methods have grabbed
much attention from the research community in the field of information security. Information
hiding refers to embedding secret message into a digital medium [18]. The message is usually
secret text, any image containing information, a secret audio message or any object which can
have binary representation. The aim is hiding the sensitive data by embedding in an unsuspected
object. This object can be any of several formats including image, audio, digital video, or some
other digital carrier that can be used to hide secret information [19]. It is then referred to as cover
image, cover audio, and cover video respectively. Steganography, watermarking and
cryptography are considered to be information hiding techniques. Figure 2.1 shows the
categorization of information hiding techniques. Each field differs from the other depending
upon the desired characteristics and attributes.
Fig 2.1: Categorization of Information Hidding Techniques
9
Table2.1 summarizes the differences and similarities between steganography, watermarking and
cryptography.
Table 2.1: Comparison of Information Hiding techniques
Information hiding is used in different applications as follows [20]:
• Copyright Protection, Single Ownership and Joint Ownership are applications based on
watermark techniques to advocate intellectual property.
• In E-commerce, registration information can be embedded in electronic papers that can be
utilized to identify authentication which is based on steganographic techniques.
10
• In medicine, doctors can embed some sensitive information such as name, comments or
diagnoses of some particular patient into their medical images. These medical images can be of
several formats such as embedding secret information into ECG images
• In military, the content of communication and the communication medium itself between
agencies must be kept secret. Information hiding technique can be used when two or more
agencies are communicating via Digital Short Radio.
• In remote sensing, information can be concealed into some site images to grant access to the
secret data only to intended users.
2.2 STEGANOGRAPHY
Steganography is a combination of two Greek words first is stegos which means cover and the
other is Grafia referred as writing, resulting in as covered writing. This field is considered to be a
modern tool utilized to ensure secure communication [21]. Steganography is basically defined as
embedding of the secret information in a cover to ensure that existence of the hidden information
is unsusceptible. The medium used for hiding the information is usually text, image, audio or
video and they all are termed as cover object e.g. cover image. After embedding the cover
medium is termed as stego-medium. Secret key can also be used for ensuring security of
extraction procedure making it more difficult for unauthorized users to extract information [22].
The steganography was first introduced back in 440 BC. Steganography as a name was
introduced around in the last decade of 15th century; whereas in actual steganography was widely
used several centuries ago. Initially wax writing tables were used to hide data, rabbits were used
by marking the secret information on their stomach, or slave’s scalps were used to hide the
information under the hair. Some special invisible ink was also introduced for covered writing
[23].
Modern steganographic systems majorly use images, audio, videos etc for information hiding
because usually people exchange the above mentioned objects using different communication
protocols [24]. Recent steganography techniques focus on hiding information in the above
mentioned files or in IP Headers.
Steganography algorithm is based on following elements:
11
1. Cover object (C) used for hiding.
2. Data or secret information termed as (M), which can be any binary information.
3. The algorithm for embedding
4. Secret (K) can also be utilized for improving the security of embedding and extraction
process
Cover medium is the criterion to classify steganography into five categories as shown in Figure
2.2:
Fig 2.2: Types of Steganography
2.2.1 TEXT STEGANOGRAPHY
Text file as a cover for hiding information is the most commonly used steganography method.
Text steganography utilizes text file for hiding secret data. Commonly used techniques in text
steganography are line shift, hiding in paragraphs, missing letters puzzle etc.
2.2.2 IMAGE STEGANOGRAPHY
A popular choice to be used as cover medium for steganography systems is considered to be
digital images. The secret information is embedded in a cover image through an efficient
algorithm which can be based on a stego key [25][26]. After the information has been embedded
the generated stego-image is transmitted to the concerned party. Receiver employs an extraction
algorithm to extract hidden information from the stego image. It is ensured that an unauthorized
12
party should only be able to notice the communication but should not be able to detect the hidden
secret message being transmitted through images.
2.2.3 AUDIO STEGANOGRAPHY
Embedding secret information in an audio cover using an efficient embedding algorithm is
termed as audio steganography. Figure of merit for transmission of sensitive data is Security and
robustness against attacks, it should also be ensured that any intruder should not be able to access
the information [27]. Existing audio steganography algorithms use mostly WAV and MP3 sound
files for embedding secret information.
2.2.4 VIDEO STEGANOGRAPHY
Video steganography uses video as cover medium for hiding secret information. Video is
considered to be a secure medium because video frequency changes frequently and video color
varies very fast which is very hard to detect through naked eye. Video steganography algorithms
hide data in the frames of video using DCT transform [28]. Most common video steganography
techniques are tri-way pixel value differencing, embedding using motion vectors.
2.2.5 PROTOCOL STEGANOGRAPHY
The secret information is embedded by using network control protocol like http, ftp, tcp, Ssh,
udp etc. Secret information is embedded in voice-over IP. Protocol steganography is a new
dimension of steganography and is considered to be more secure than other dimensions. Modern
research in this field is focusing on improving the capacity and robustness of this category [29].
2.3 IMAGE STEGANOGRAPHY
Image steganography deals with hiding text, speech or an image within an image. Image
steganography is mainly categorized into two types depending on the embedding method used
[30].
13
1. Spatial Domain
2. Transform Domain
2.3.1 SPATIAL DOMAIN
Image steganography techniques which embed the secret information by directly modifying the
values of the cover image pixels are termed as spatial domain techniques. The focus of spatial
domain techniques is improving the capacity and imperceptibility of the system [31]. The most
commonly used technique in this category is LSB steganography methodology based on
replacement of the LSBs of cover image pixels by the information bits [32]. Many techniques
have been discussed in literature to improve the capacity of spatial domain techniques.
2.3.2 TRANSFORM DOMAIN
These techniques work by first converting the cover image is to frequency domain by applying
some transformation. The secret information is embedded by mostly utilizing LSB embedding
but in this case the coefficients calculated against the cover image are used and after embedding
the cover image is applied with the inverse transformation to generate stego image in spatial
domain [33]. Commonly used transforms in this regard are DCT and DWT. These techniques are
proven to perform better than spatial domain techniques in terms of robustness [34][35].
2.4 RELATED WORK
Raftari, N and Moghadam [36] presented an algorithm as a combination of and discrete cosine
transform (DCT) and the integer wavelet transform. Munkres' assignment algorithm is applied
for embedding the secret information. Prabakaran et al., [37] presented a steganography
approach using DWT transform to represent the secret and cover images in frequency domain
and embedding coefficients of secret message into transform domain version of the cover
Motamedi, H. [38] used DWT in image steganography to utilize algorithms based on denoising.
Steganography generally replaces the noise components of cover image for information hiding.
Spatial domain techniques usually work by changing the pixel values of the cover image in order
to embed secret data. The criterion to measure embedding rate is defined as bit per pixel (bpp) in
image steganography applications. Ioannidou, A et al., [39] used a methodology based on hiding
larger payload in cover image areas that consist of sharper edges. The focus of presented
14
technique is to utilize the sharp edges in the cover image for data hiding. The proposed method
was not able to enhance the payload capacity in the images having relatively smoother areas.
Hong, W., et al, [40] El-Emam, N. and Al-Zubidy, R[41] presented a scheme for enhancing
payload capacity by increasing security layers to four times. This work presents an image
segmentation algorithm by combining adaptive neural networks with GA resulting into an
intelligent combination. The presented technique has a large computation time to generate a
stego image with high imperceptibility. Hemalatha et al, [42] used a method for hiding two
secret images in single cover while maintaining the PSNR of stego image. But the
imperceptibility was not up to the mark due to increased payload.. Li, Y. et al [43], presented an
algorithm based on Adjacent Pixel Difference (APD), which improves the payload capacity by
employing histogram of the pixel difference. The algorithm is designed for grayscale images, but
as a figure of merit PSNR is not enough to ensure imperceptibility and also, the system is not
tested against attacks. Zhu, Y et al., [44] provided an algorithm without utilizing any assumption
and it was theoretically proved the security of the proposed algorithm is remarkable against
multiple attacks. Wang et al, [45] proposed scheme termed as reversible spatial domain
technique, histogram shifting is the key feature. The main achievement was increase in payload
capacity. However, there was no focus on testing the algorithm against different image
processing attacks.
K.Parametha et al., [46] proposed an algorithm that is based on selection of 2×2 blocks and the
selection criterion is contrast of the particular block. The selected high contrast blocks are used
for embedding secret data by utilizing Mod-4 embedding method. The proposed embedding
method to minimize the modifications caused in cover image while embedding. Embedding
capacity is measured as the amount of information in bits that can be embedded in a particular
cover medium. Proposed algorithm claimed increase in embedding capacity while maintaining
imperceptibility of the stego images.
Sara sajasi et al., [47] presented an innovative a steganography algorithm for embedding a secret
image within a gray-scale image for the enhancement in imperceptibility of the system while
providing reasonable capacity. The main feature is the integration of Fuzzy Inference System
(FIS) and the conventional HVS system. As it is known that modifications in the parts of an
image which are relatively rough are hardly detectable, texture, edge, and brightness are the three
15
parameters evaluated against each sub part of cover. These features act as the input for FIS in
order to calculate capacity of each all the blocks of cover image. The classification of blocks is
done resulting into five different types depending on the information hiding capability of each
block. LSB embedding method is used for hiding secret information in all blocks and the
information is embedded according to the payload capacity calculated by FIS.
Mansi S. Subhedar et al. tried to ensure reliable transmission of information in [48] when the
steganalysis is carried out by adopting a technique to smartly select a cover image and the
selection is done depending on the measurement of the contrast of cover. The selected cover is
used for embedding information in the frequency domain using contourlet transform. The
selection criterion and counterlet transform resulted in improved imperceptibility of the system
ensuring secure data transmission. The proposed algorithm is supported by presented simulation
results showing high performance when assessed by steganography metrics.
A Discrete Wavelet Transform (DWT) based image steganography algorithm for perfect security
and high payload capacity is proposed by Mohammad Reza Dastjani Farahani and Ali
Pourmohammad in [49]. The algorithm hides image within a cover image. both the cover and
message image are converted to frequency domain by applying DWT and the coefficients
computed against the message are used as secret message to be embedded in the coefficients
related to the cover image. Various secret messages are embedded and the algorithm has been
proved as robust for all the cases. The performance is analyzed using imperceptibility and
robustness as quality measurement parameters. Results emphasize that the algorithm enhances
imperceptibility and robustness.
Bingwen Feng et al., [50] demonstrated an algorithms focusing on minimization of the artifacts
produced in a particular cover image due to hiding the binary images. The scheme is based on
extraction of texture based parameters from the binary image. The modification in each pixel is
considered to find the average modification. By testing on both simple binary images and the
constructed image data set, it has been shown that the proposed algorithm was able to identify
distortions and using that information improved imperceptibility. Using these measurements, an
algorithm with practical considerations is developed. The generation of cover image is done by
finding out some super pixels offering high capacity. Syndrome-trellis code has been utilized for
ensuring minimum distortive effects. Simulations show that the algorithm improves security,
payload capacity and imperceptibility is also up to the mark.
16
Considering the security of LSB embedding Vikas Verma et al., [51] embedded information in
LSBs of a RGB image. The embedding is done in a circular fashion. The proposed method
calculates the central pixel and radius of the circle calculated against height dimensions of the
image. Central pixel is the starting point for embedding the information and the process
continues in clockwise direction.
Bin Li et al., [52] presented a method that tries to minimize the distortions by using the
modification relationships. Presented scheme utilizes a novel strategy (CMDs), which assumes
that when information is embedded in heavy textured part of the image the direction of
modification is in same or one particular direction. The algorithm works by first dividing the
cover image into sub images. In each divided part the information is embedded and it tried that
the modifications in a selected reference pixel and its surrounding pixel are in same direction.
Algorithm is proved to be secure against the modern stego attacks and results are well presented
to support the claim.
Saiful Islam et al., [53] proposed steganography method, where embedding is done in edges of
the cover image. The selection criterion for embedding capacity has been chosen as weaker
edges are used for large payload embedding. Presented results indicate that the proposed
technique enhances the capacity while works similar in terms of security when compared with
the available methods.
Y. K. Jain et al., [54] have shown an adaptive LSB embedding method. The proposed method
divides the 8-bit format image pixels ranging between (0-255) and generates a Stego-key based
on this computation. This generated stego-key assigns five gray level ranges to the related cover
image and out of generated ranges every range decides the number of bits that can be embedded
in LSBs of the cover image. The major contribution of this algorithm is the huge improvement in
the secrecy of information and high embedding strength. The algorithm has a shortcoming that is
some extra payload has to be embedded to improve integrity of the system. Authors also
proposed a method for RGB images by only modifying the blue channel using the proposed
scheme for information hiding. The algorithm focused on achieving high capacity and security.
Yang et al., in [55] presented a work which utilizes a scheme for embedding secret information
in cover image based on adaptive LSB substitution. The noise sensitive area of the image has
been used for embedding for improving imperceptibility. Proposed method extracts and utilizes
17
the image areas with normal texture and the areas containing edges in order to embed data.
Different parameters including texture, brightness and edges are extracted to predict the capacity
of each area of the cover image. The embedding rate is high at the areas which are less sensitive
and it would be low in the regions considered to be more sensitive in order to balance overall
imperceptibility of the cover image. The algorithm also incorporates the pixel adjustment method
that helped in improvement of stego-image visual quality.
S. Channalli and A. Jadhav [56] introduced a data hiding scheme utilizing LSB embedding. To
generate a security key algorithm used common pattern bits. Data embedding in the pixels is
dependent on the stego key and the information bits. The process of embedding works on the
comparison of pattern and the information bit. On a true comparison the 2nd bit of the pixel is
modified and on a false comparison no modification is made. The technique tried to achieve
security of the system but the cost paid is in terms of high computations required for each pixel.
C.-H. Yang et al., [57] proposed a Pixel value difference (PVD) and LSB scheme to achieve
adaptive LSB data embedding method. In pixel value differencing (PVD) the capacity in a
particular pair of pixels is dependent on the difference between the corresponding pair. PVD
method is generally considered to provide high imperceptibility; the calculated capacity is
dependent on pixel difference. The areas containing edges have been used as regions for high
embedding rate while smoother regions are termed as areas offering low capacity. So the
proposed technique provides larger payload capacity and high imperceptibility as indicated by
experimental results. But one can say the complexity of method is relatively high.
K.-H. Jung et al., [58] have utilized a method called MPD which used a block of four pixels
instead of using a pair. The difference between the block elements is calculated and the blocks
resulting in smaller difference employ LSB embedding while the larger difference blocks used
MPD for data hiding. Algorithm is simple but it has been tested for limited cases.
H. Zhang et al., [59] presented a technique which also used PVD; but in this work three pixels
are used instead of a pair for embedding information. The proposed method uses simple LSB
embedding method for secret data hiding. The capacity of each block is computed by setting a
reference pixel and the difference of the reference from the three neighboring pixels is computed.
Data is embedded in the blocks which offer high difference value using LSB embedding. The
proposed method employed adaptive embedding method resulting in better capacity and
18
imperceptibility. Advantage of proposed method is minimum change in the histogram of the
cover image which makes statistical detection difficult.
W. J. Chen et al., [60] have introduced a steganography algorithm combining edge detection
scheme and the conventional embedding based on LSB modification. For detecting the regions
of image containing more edges a combination of canny and FIS has been utilized. The
incorporation of edge detection improved the algorithm in terms of imperceptibility when
compared with simple LSB embedding. The proposed algorithm needs to be tested on more
images to support the claimed results.
Madhu et al., in [61] proposed an image steganography method, which is based on LSB
substitution and random pixel selection of cover image. The method targets on improving the
security by adding password to LSB of pixels. The region of interest to hide information is
selected using random number generation. The algorithm has improved security, but did not
consider the imperceptibility factor.
M. A. Al-Husainy [62] proposed an algorithm which utilizes the English letters and some are the
fixed characters total 32 in number for mapping of the cover pixels. A tabular form is generated
representing four different cases and the pixels are mapped by using high probability of
matching. The embedding is done in 7 bits. The drawback is algorithm requires huge memory for
retaining the mapping data. The algorithm needs this information for data extraction. The
algorithm is only feasible for text embedding applications.
H. Motameni et al., [63], proposed an algorithm in which the embedding is done by first
identifying the cover image regions which are darker in nature. Embedding is based on
conventional LSB data embedding method. The cover image is converted to binary scheme
based on exploiting 8-pixel connectivity. High computations are required for identifying the
darker region and the connectivity of the pixels. The capacity depends entirely on the texture of
image.
M. Tanvir Parvez et al., [64] have introduced an algorithm based on pixel indication utilizing
adaptive bits; the proposed scheme uses one component out of RGB components and the
embedding rate is variable depending on the intensity level of the selected pixel. Proposed
method provides minimum histogram difference due to modification in the cover image.
Hamid et al., in [65] have worked on image steganography algorithm based on classification of
the cover image regions. The regions of cover image are classified into complex and simple
19
categories using texture analysis. The information in simple regions is embedded in “3-3-2”
fashion, which means if we consider 24-bit format 3 bits are embedded in red, three in green and
two in the blue component. In complex regions 4 bits are embedded in every component. The
combination of two different embedding methods based on texture classification offered high
imperceptibility while the algorithm also offered enhancement in capacity as well.
.
M. Chaumont et al., in [66] worked on an algorithm that utilizes DCT based transform for
embedding the information. A compressed gray-level image has been used for hiding color
information. The proposed algorithm is a merger of data hiding combined with the quantization
and sorting of colors. The proposed method provides free access to gray-level image to but
restricted access is provided for the same color images. The information is accessible to the
legitimate user possessing stego key.
K. S. Babu et al., in [67] proposed a scheme for using images to hide the secret information. The
algorithm also has a mechanism to verify the integrity of the information at receiver side. The
algorithm computes coefficients related to the original information by applying DWT transform
and these calculated coefficients are sorted by using a secret verification code. The newly
arranged coefficients are embedded and the verification code helps in retrieval at the decoder.
This code also helps in minimizing the chance of tempering in the secret data.
Mamta Juneja et al., [68] proposed a robust technique based on RSA encryption and LSB
embedding. The algorithm works on selecting a best and suitable cover image for some
particular message. The selection is based on finding a cover having similarity with the
information to be hidden which makes the presence of secret information undetectable. RSA
encryption standard has been used for encrypt the information prior to embedding. This method
proved to be superior as compared to the current steganography tools.
Subba Rao Y.V et al., [69] presented a work that utilizes random sorting of the encrypted
message bits before embedding them into the cover. Different sorted sequences are compared
with the cover image and a sequence which is matched closest to the cover image is chosen for
embedding. This random sorting is done by applying L.F.S.R (Linear Feedback Shift Register).
A considerable advantage of this technique is the removal of (1-1) mapping between encrypted
information and cover medium. Secondly, for attacker to carry a successful cryptanalysis, the
information of the L.F.S.R is needed which has been used to generate the random sequence.
20
Septimiu Fabian Mare, Et al., [70] proposed a new image steganography algorithm which
incorporates two well known encryption methods that are AES and RSA. The secret information
is given a cover of dual encryption making it more robust against cryptanalysis and other attacks.
The information is first encrypted by using AES followed by RSA. To generate secrete key hash
function is computed against the related cover image which is not affected by embedding process
making the system more robust and secure. The combination promises high security and integrity
of the secret information.
S Usha et al., [71] presented a technique which also combines steganography and encryption.
The authors have used double encryption for the secret message. Firstly message is applied with
playfair cipher technique and the encrypted information is provided with the second level of
security by using advance encryption standard (AES). The double encrypted message is
embedded in the three components of RGB cover image using LSB embedding. The presented
results prove that the proposed model is secure.
Yang Ren-er et al., [72] have focused on security improvement of steganography and this has
been achieved by using a blend of DES and steganography. The algorithm works on
preprocessing the secret information and this added module encrypts the secret information using
DES. It is shown that the incorporation of DES provides considerable improvement in the anti
detection property of the system. Results presented aid to the claim made that the system has a
better performance in terms of anti-detection and robustness.
Diaa E.M.Ahmed et al., [73] presented a work which integrates elliptic curve cryptography
(ECC) to image steganography. The secret text message is transformed to binary form and is
encrypted to cipher text by applying ECC. Calculation of all the parameters of ECC has been
done and shown in the paper. This encrypted information is then embedded to cover image.
Results presented show the system is robust against statistical attacks.
Dipanwita Debnath et al [74], presented a steganography method designed in spatial domain.
Authors have proposed a new mapping rule. The algorithm is based on first utilizing some newly
defined mapping tables for conversion of secret information to text and then applying hill cipher
technique. The encrypted information is embedded into the three channels of a RGB cover
image. The mapping rule and encryption increase the security while embedding in three channels
21
enhances the capacity. Results are well presented to show that claimed objectives of the
proposed work have been successfully achieved.
Jeng-Shyang Pan et al., [75] Proposed image steganography algorithm which is a blend of
compressed sensing and steganography, the algorithm is capable of hiding the data in an
innovative domain. Using the prior knowledge that most of the natural images can be
compressed in a known transform domain the attributes of CS have been used. The random
generation of measurement basis acts as a secret key. The reconstruction of stego image is done
by using total variation minimization algorithm. The proposed model is considered to be accurate
in recovery and highly secure
Nitin Kaul et al [76] presented an algorithm for embedding an audio message in an image. The
audio is compressed using wavelet transform and LSB embedding for hiding compressed audio
message into the cover image. Amol Bhujade et al [77] used RGB image to embed binary audio
data. The last two significant bits of Red, Green and Blue pixel components are used to hide data
so that each pixel can hide 6 bits of information.
Devendra Singh Rao et al [78] presented audio embedding in RGB components of the cover
image. The selected pixels for embedding information are based on a circle equation where the
central pixel of the circle and radius of the circle for embedding information are considered as
secret key of the system. RAMESH GOTTIPATI et al [79] presented the scheme for hiding an
audio in an image. The proposed algorithm is another application of the combination of
steganography and encryption The audio message is encrypted by applying AES encryption and
the resultant information is embedded in an image using conventional LSB embedding method.
SUMMARY
This chapter reviewed and compared basic techniques that are being used in information
security. Steganography has been discussed and different types of steganography algorithms are
presented. Image steganography and the related work presenting the techniques used in this field
have been discussed in detail.
22
Chapter 3
MATHEMATICAL
TECHNIQUES
3.1 INTRODUCTION
This chapter is based on different mathematical techniques being utilized in the presented
research. We would be presenting three main mathematical techniques, compressed sensing,
substitution boxes and chaotic equations.
The distribution of the remaining chapter is as follows. The details of compressed sensing are
presented in Section 3.2. Section 3.3 presents an overview of S-Boxes and Section 3.4 presents
chaotic systems.
3.2 COMPRESSED SENSING
Compressed sensing was initially introduced in 2006 by two ground breaking research papers,
which were [80] by Donoho and [81] by Cand´es, Romberg, and Tao. If we explore the recent
literature, compressed sensing has been widely applied in many research areas. This innovative
technique is extensively used in mathematics, signal processing, MRI, biology, radar
communication, and many other applications [82]. Compressed sensing is a method which is
able to reconstruct a sparse signal with very few measurements by utilizing smart optimization
algorithms [83]. Considering another approach, the technique ensures an exact recovery of the
signal after it has been significantly reduced in dimensions. We can also say the system is an
application of overcomplete system [84], [85]. Compressed sensing has it theoretical foundation
on variety of methodologies including optimization algorithms, mathematical transformation
analysis and signal processing [86].
23
Now let us see this problem mathematically. Let the signal of interest be fnx1. If we project this
signal on measurement basis Φmxn, where m < n, this operation yields:
yk = <f, Φk> for k = 1, 2. …, m
So knowing m projections i.e �������� and basis Φmxn , can we find f:
ymx1 = Φmxn fnx1
It is not possible to have a unique solution if we use the knowledge of linear algebra as
unknowns (n) are more than the number of equations (m). The answer is given by compressed
sensing theory. The recovery is possible if following two conditions are satisfied [87].
1. Sparsity
2. Incoherence
3.2.1 SPARSITY
As a first consideration, one can ask whether sparsity is a realistic assumption [88]. We consider
an example of an image when transformed using DWT. It can be clearly seen that most of the
coefficients are minimum in magnitude; the darker color in Figure 3.1 indicates these
coefficients [89].
Fig 3.1: Sparsity in Images
24
Various representation systems are devised which are chosen depending on the signal of interest.
Recent research indicated that DWT based systems cannot provide optimal sparse representation
for most of the natural images, but shearlets which has been recently devised is a better option
[90], [91]. Hence, if we assume some foreknowledge of the signal a suitable and mathematically
stable representation system should be known. For most of the considered applications, the
signal “x” is usually sparse, best example is any natural images. Thus, x being sparse itself is a
natural assumption [92].
If we discuss sparsity in mathematical terms for a signal “f” there exist some representation basis
����� so that
� � ∑ ���� �3.1�
Number of nonzeros zi is s << n, so most of the coefficients (n-s) zi are zero. We call f as s-sparse
in ψ.
3.2.2 INCOHERENT SAMPLING
Let us have a pair of sensing basis Φ of order m x n and sparse representation basis ψ of order n
x n which we need to use in compressed sensing problem. The mutual coherence between a
given pair (Φ,ψ) is defined as:
The parameter “µ” quantifies the correlation between the elements of Φ and ψ which lies in the
range [1, √n] [93]. Compressive sensing requires low coherence. A pair of basis is considered to
be ideal to be utilized in CS if the coherence is small between a given pair of basis. Random
matrices when used as sensing basis are proven to be largely incoherent with the known sparse
representation basis [94].
Recovery :
The knowledge of “y” is used to recover a sparse vector x by using the following relationship:
(3.2)
25
������� �. � � � �� � �� � �� � !"! � � � (3.3)
As “l0-NORM” is a non convex norm the problem is considered to be NP-hard. So we instead
use ‘‘l1-NORM’ resulting into the new problem,
min& ���� �. � � � �� �3.4�
At this point we need to mention a case when the signal is not exactly sparse i.e. the coefficients
calculated against the signal using representation basis are nearly zero. Moreover there is some
measurement noise in the system with variance ε [95][96]. In this case another necessary
condition to be met for the reconstruction of the signal that is explained below:
3.2.3 RESTRICTED ISOMETRY PROPERTY
This parameter is considered to be directly associated with the robustness of the compressed
sensing [97][98]. The restricted isometry constant of order “s” δs = δs(A) corresponding to
matrix A of order m x n is the minimum possible δ ≥ 0 such that:
If this property holds the matrix A is considered to be a transformation that preserves the
Euclidian length of S-sparse signal. This ensures that the S-sparse vectors do not lie in the null
space of matrix A which if not ensured would make the recovery impossible [99], [100].
Recovery:
In this case the recovery algorithm is updated to following equation:
Where ε is noise variance. It can be transformed into unconstrained lagrangian problem.
min&�( ��� ) *�(( + ,���� (3.7)
(3.5)
(3.6)
26
The overall CS model is shown in figure 3.2.
Fig3.2: Compressed Sensing Model
3.2.4 RECOVERY ALGORITHMS
Recovery of the compressed information is considered to be the most critical and important step
in compressed sensing. The recovery algorithm needs to be designed in such a way that it can
attain an optimum solution while minimizing the computational time and complexity. The most
utilized algorithms are summarized in Fig. 3.3 [101], [102].
Fig 3.3: Recovery Algorithms
Recovery Algorithms
Convex Optimization
Greedy Algorithms
Hard Thresholding
Soft Thresholding
27
We have used convex optimization for the recovery which has already been explained in section
3.2.
3.3 SUBSTITUTION-BOXES
The fundamental concepts of confusion and diffusion as identified by Shannon are the
foundations of any cipher system. The S-boxes are an integral part of symmetric key
cryptosystems [103]. Their basic purpose is to provide the necessary confusion. They are used
for obscuring the relation between the plain text and the cipher text. They are essentially non-
linear mapping which take input of a certain number of bits and converts it into some different
bits. The number of bits at input and output need not be equal. The security of systems using the
S-boxes depends a on their proper selection. A good amount of research has been going on in
this direction [104].
S-boxes are designed to resist different attacks that are launched by cryptanalysis techniques for
breaking the security of the system. The conditions defined for a robust design of S-box are:
1. The Strict Avalanche Criteria (SAC) should be strictly met
2. Non-linearity
3. Robustness against cryptanalysis
SAC is a measure of changes produced in the encrypted sequence when the input is modified. “If
a cryptographic function is to satisfy the strict avalanche criterion, then each output bit should
change with a probability of one half whenever a single input bit is complemented.”
Non-linearity is considered to be the attribute of the component function of S-Box. The
component function should be non linear. These two properties ensure the high resistance of S
Box against cryptanalysis [105].
Figure 3.4 shows an example of DES S-box.
28
Fig 3.4: An Example of S-Box
In the presented example the input is of 6-bits, the output is of 4-bits. The outer two bits are used
for row selection, and the inner four bits are utilized for column selection. It has been shown in
the current literature that a small modification in a substitution box can lead to declining of its
robustness [106].
3.4 CHAOTIC KEY GENERATION
This section provides an overview of chaotic number generation based on chaotic systems.
3.4.1 CHAOTIC SYSTEMS
Chaotic Systems are considered to be dynamical systems with nonlinear attributes. Difference
equations for continuous case and differential equations are utilized to represent these systems in
continuous domain. Initial conditions play a vital role in chaotic systems as exactly same initial
conditions are required to produce two identical chaotic sequences.
3.4.2 THE LOGISTIC MAP SYSTEM
Following equation describes the logistic map system:
The current state of the system is xn+1, xn is considered as the previous state and the constant r
ranges between 2 to 4. Any small modification in x0 or r can result into a totally different chaotic
(3.8)
29
sequence, so to ensure the exact sequence generation at receiver if used in any communication
system the initial conditions must be exactly same [107]-[115].
3.4.3 UTILIZING LOGISTIC MAP SYSTEM IN IMAGE STEGANOGRAPHY
The logistic map as explained generates a random sequence that is based on provided initial
conditions. This phenomenon has been utilized to randomly select the locations of cover image
pixels which are to be utilized for embedding secret information. The process is described by
following steps:
1. Generate a binary chaotic sequence using the equation :
2. Multiply the generated sequences by 8 and take its ceil(.) so the generated sequence
converts into a range between 1 to 8.
3. Let we define the generated chaotic sequence as:
S1 = X1, X2, X3, X4, X5,…
4. Let we define a new sequence based on S1:
S2 = X1, X1+X2, X1+X2+X3, X1+X2+X3+X4,….
This generated sequence is an increasing sequence. Each entry of this sequence is used to
indicate the location of the cover image pixel in which the information would be embedded. Let
us see this with the help of an example:
Example:
Let the generated sequence by chaotic map be:
S1 = 1 3 5 2 7 3 6 1 3…
30
Than S2 would be:
S2 = 1, 4, 9, 11, 18, 21, 27, 28, 31…
So the location of first pixel for embedding information is 1, second pixel to be used for
embedding would be 4th, next location for embedding would be 9th pixel and so on.
31
Chapter 4
IMPROVING SECURITY, CAPACITY
AND IMPERCEPTIBILITY OF IMAGE
STEGANOGRAPHY BY COMBINING
ENCRYPTION, ERROR CORRECTION
CODES AND PRIME SERIES
REPRESENTATION WITH
STEGANOGRAPHY
4.1 INTRODUCTION
This chapter focuses on two different methodologies we have proposed for secure
communication using image steganography. The first proposed secure communication problem
is based on encrypting secrete binary text information using right translated AES gray S-boxes
presented by Mubashar Khan et al in [116] combined with image steganography. Furthermore
the binary text information which is the secret message is given a cover of BCH error correction
codes to ensure accurate detection. The pre processed secret information is embedded in
chaotically selected pixels whose selection is optimized by GA as discussed in [117] using newly
proposed differential LSB embedding. The combination of encryption, error correction code,
optimized chaotically embedding through differential LSB encoding make the secret
transmission of data incomprehensible and invisible. This combination also enhances the anti
detection performance of system and increases the PSNR of cover image.
The second proposed technique is a novel embedding method that is based on prime series
representation of cover image pixels that have been utilized for hiding the secret information.
The proposed prime series representation converts the pixel into 13-bit format providing more bit
32
planes to embed information as compared to the conventional 8-bit binary representation. We
have used 3 least significant bit plans to embed secret information resultantly providing three
times increase in capacity. To ensure the complete embed and random spread of the secret
information into the whole cover, we have proposed an adaptive chaotic key generation
algorithm. The algorithm adjusts the chaotic key which randomly selects the embedding pixels
based on the cover image and the secret image such that all the secret information is embedded
in the cover image and at the same time, it is spread all over the cover image. Redundancy in the
message image has been removed by applying 2-D DCT and thresholding of the coefficients.
After thresholding the new dimensions of the secret image are calculated by locating the row and
column that contains the last non zero coefficient. Only the block information of the calculated
reduced size is kept for embedding. Thus we reduce the size of the payload and need to send the
calculated row and column number rather than sending the locations of all the non zero
coefficients. This results in reduction of large overhead information thus enhancing the capacity
of the system. The coefficients are given a cover of 2-bit error correction Reed Solomon code to
ensure reliable recovery of the information.
The rest of the chapter is organized as follows. Section 4.2 discusses the first proposed method,
its modules and achieved results in detail. Section 4.3 is based on the second contribution
presenting its each and every detail and the achieved results and discussion. Section 4.4
summarizes the chapter.
4.2 IMPROVED AND SECURE DIFFERENTIAL LSB
EMBEDDING STEGANOGRAPHY BASED ON CHAOS,
GENETIC ALGORITHM AND BCH CODES
4.2.1 PROPOSED MODEL
The overall idea of proposed method is shown in following figure:
33
Fig 4.1: Proposed Method
The proposed work can be described in two modules.
1) Secret information pre-processing
2) Embedding processed information to cover image
1) Secret Information Pre-Processing
The binary information to be transmitted is pre-processed to increase the non-comprehensible
feature of the information. The information processing module is a combination of two sub
modules.
Chaotically selected
translated SBOX
Stego Object
2-bit error correcting code
(15,7)BCH encoder
Differential LSB embedding
in chaotically selected pixels
Secrete Binary
Information
Cover Image
Secret Key 1
Optimized Secret Key 2
by GA
34
a) Right translated gray box
b) BCH channel encoder
a) Right translated gray S-box
This idea is presented by Mubashir khan et al [116]. The devised S-box is an application
of right translation followed by gray codes over the AES S-Box which is widely used.
The proposed substitution box in [116] consists of three transformations
1. AES S-Box
2. Right translation
3. Gray Code
Fig 4.2: Right Translated Gray S-Box
The AES S-Box is a bijective transformation applied on 8-bit data defined as:
. SAES (x) = - . �� � � 00 ° 2��� !3�!� !"!4 (4.1)
Where b = [1 1 0 0 0 1 1 0]T and x is the binary input. 2 is applied as a group
automorphism on elements of x defined as 2 = x-1. 0 is defined as: 0 (y) = My + b.
considered to be affine transformation being applied on y. Where
XOR
AES S-Box
Gray Code
Encoder
8 bit Binary Information
Chaotically generated binary 8 bit vector
8 bit Encrypted
Information
35
M =
566666671 0 0 0 1 1 1 11 1 0 0 0 1 1 11 1 1 0 0 0 1 11 1 1 1 0 0 0 11 1 1 1 1 0 0 00 1 1 1 1 1 0 00 0 1 1 1 1 1 00 0 0 1 1 1 1 18
999999:
Right translation is applied to the binary output of AES S-Box. Right translation is done
by chaotically generating a vector and this vector is xored with the AES binary output.
This adds on to the security of the system as this chaotically generated random vector is
based on information known to the intended user only. The generation is same as
discussed in Chapter 3. Finally the gray code is calculated against the translated output
using relation:
gn = -1 ) .� �� .�;� � 1.� <� !"���! 4
(4.2)
Overall representation of right translated S-Box is given by:
ζ(g) = SAES ° ρg ° G (4.3)
Simulation results presented against this S-Box in [116] indicate that the proposed S-Box
has better resistance against computational attacks and The algorithm is designed so
carefully that all newly generated S-boxes preserve all the cryptographically important
properties including nonlinearity, bit independence and strict avalanche, linear
approximation and differential approximation of the original AES S-box.
b) Channel encoder
Error correction codes are widely utilized in communication to ensure error free data
transmission. There are different categories being utilized like block codes, cyclic codes,
convolution codes, reed solomon codes and BCH codes. All above mentioned techniques
are categorized as forward error correction code in which decoder has the capability to
36
detect and correct errors. The encoder is provided with k-bits data corresponding to
which it generates n bit code where n is greater than k. The (n-k) redundant bits are
utilized to correct error. We have utilized BCH codes in this paper.
The binary output of the translated S-Box is passed to (15,7) BCH channel encoder. The
encoder divides the binary bits provided to 7-bit chunks and generates an output of 15
bits corresponding to each 7 bit chunk. The code can correct up to two bit error.
2) Embedding Method
The embedding method includes LSB based image steganography combined with chaotic
pixel selection and differential encoding. So embedding method is described by using
two sub modules.
a) Chaotic Pixel Selection
b) Differential Encoder
a) Optimized Chaotic pixel selection
This module chaotically locates the pixels which are to be used for information
embedding. The chaotic selection is done by using following equation.
Where “r” is a bifurcation parameter and for system to be chaotic and r lies between 3.57
and 4 and xo is an initial condition having value between 0 and 1. Chaotic systems exhibit
numerous characteristics that distinguish it from the other systems. One feature is its
sensitivity to the initial conditions which if slightly modified; an entirely different pattern
is obtained. So to generate the exact chaotic sequence exact information about xo and r is
compulsory. An array of chaotic integers is generated between 1 to M x N. M x N is the
size of cover image. These integers are basically the positions of pixels which are utilized
to embed information. This selection is optimized by using huristic optimization
algorithm GA. The aim of using GA is to chaotically select the pixels in a way to
minimize MSE between cover image and stego image. The scheme in return maximized
PSNR of the stego image making the secret message more secure and invisible. Figure 3
(4.4)
37
illustrates the flow diagram of stepwise optimization algorithm. When the pixels have
been chaotically located their values are converted to 8-bit binary format.
38
Fig 4.3 Optimized Chaotic Key Generation using GA
Next generation of (r,xo)1 , (r,xo)2
….(r,xo)N
Start
(r,xo)1 (r,xo)2 (r,xo)N
Initialization
Chaotic Key generator
Key 1 Key 2 Key N
Encrypted Message
Differential LSB Embedding
Differential LSB Embedding
Differential LSB Embedding
Fitness 1 Fitness 2 Fitness N
GA Operators
No of generations> maxGen
Best Solution
End
39
b) Differential LSB encoder
The advantage of differential LSB is that the information is not conveyed by the absolute
phase of the signal with respect to a reference phase but by difference of phases between
successive symbols. This eliminates the phase reference at receiver and avoids error
propagation. Following figure shows the detail of differential encoding. The information
is embedded to the LSBs of chaotically chosen pixels.
Fig 4.4 Differential LSB Encoder
The embedding in the least significant bit is shown by following example. Let the chaotically
selected pixel is pixel number 1 and its value is 12. Let the information bit to be embedded after
differential encoding be “1”. The selected pixel is converted to 8-bit binary format as shown
below:
Bit
Weight
1 2 4 8 16 32 64 128
Value 0 0 1 1 0 0 0 0
Is
inormation
bit == 1
Inverter
Previously embedded bit
Current information bit to be embedded
Current bit to be embedded =
previous embedded bit
Current bit to be embedded = (previously
embedded bit)’
Yes
No
40
The LSB of the pixel value is “0” in this case that is replaced by the information bit which
resultantly changes the pixel value to 13.
Bit
Weight
1 2 4 8 16 32 64 128
Value 1 0 1 1 0 0 0 0
Decoding of the proposed algorithm is exactly opposite to encoding procedure as shown in figure
4.5.
Fig 4.5: Proposed Decoder
4.2.2 SIMULATION RESULTS
The proposed algorithm was tested using three different cover images. Figure 4.6(a) shows the
original flower image and Figure 4.6(b) shows the image after embedding. Figure 4.7(a) shows
Stego Object
Data Extraction from LSBs of chaotically selected pixels and
differential decoding
Binary Information
Secret Key 2
Secret Key 1
2-bit error correcting code
(15,7)BCH decoder
Chaotically selected
translated SBOX inverse
41
the original baboon image and Figure 4.7(b) shows the image after embedding. Figure 4.8(a)
shows the original tree image and Figure 4.8(b) shows the image after embedding. Figure 4.9(a)
shows the original man image and Figure 4.9(b) shows the image after embedding. Table 4.1
shows the results achieved using the proposed model for four different cover images without
using GA optimization. Table 4.2 shows the results achieved using the proposed model for four
different cover images using GA optimization. Simulation results show low MSE between cover
image and stego image. Moreover high value of PSNR further elaborates high imperceptibility.
Comparison of table 1 and table 2 elaborates the significance of using genetic algorithm
depicting improved PSNR and minimum MSE. Following equations describe the MSE and
PSNR.
=>? � ∑ @AB�C,E�;AF�C,E�GH,ICJE (4.5)
I1 is original whereas I2 is stego image. M and N are number of rows and columns in input image
respectively.
K>LM � 103<N��OF
CPQ (4.6)
R is the maximum value a pixel can assume which is 255 in un-signed 8-bit integer data type..
Fig 4.6(a) Fig 4.6(b)
42
Figu 4.8(a) Fig 4.8(b)
Fig 4.7(b) Fig 4.7(a)
Fig 4.9(b) Fig 4.9(a)
43
Table 4.1: PSNR of different cover images without using GA optimization
Cover
Image
Cover
Image
Size
No of
Secrete
Information
Bits
No of Bits
Embedded
After Pre
Processing
MSE PSNR
dB
Flower 192 X 262 112 240 0.0023 74.22
Baboon 138 X 138 112 240 0.0060 70
Tree 192 X 262 112 240 0.0024 74
Man 209 X 153 112 240 0.0040 72.11
Table 4.2: PSNR of different cover images using GA optimization
Cover
Image
Cover
Image
Size
No of
Secrete
Information
Bits
No of Bits
Embedded
After Pre
Processing
MSE PSNR
Flower 192 X 262 112 240 0.00011 87.71
Baboon 138 X 138 112 240 0.00031 83.21
Tree 192 X 262 112 240 0.00013 86.99
Man 209 X 153 112 240 0.00018 85.57
4.3 HIGH CAPACITY IMAGE STEGANOGRAPHY BASED ON
PRIME SERIES REPRESENTATION AND PAYLOAD
REDUNDANCY REMOVAL
4.3.1 PROPOSED MODEL
One of the main goals of steganography is enhancement in the capacity, while maintaining a
certain level of imperceptibility [118]. Keeping this as target our proposed model is shown in
Figure 4.10.
44
Fig 4.10: Block Diagram of Proposed Model
The milestones set for our research are:
1. Proposing some embedding technique to enhance Embedding Capacity, removing
redundancy in the message and minimize distortion in cover image
2. Proposing an algorithm to randomly spread the message image into the entire cover
image and the algorithm adjusts this spreading depending on the cover and the message
images and some other factors
3. Making the message more secure for error free recovery
We present an innovative prime series representation of the cover image pixels in which we have
utilized prime numbers to enhance capacity by embedding 3 bits of information per pixel thus
increasing overall capacity of cover image three times. Our proposed adaptive key generator
algorithm gives approximate density spread of the message over the whole cover image. The
algorithm calculates the average distance between the cover image pixels which are chosen by
the random chaotic key such that the entire message is embedded fully into the cover image.
Furthermore the redundancy in secret message is removed using 2D-DCT transform along with
thresholding of the coefficients secured using channel codes to contribute towards improved
capacity and security. We discuss the overall proposed algorithm in subsequent sections.
Message Image
Cover Image
Proposed13-bit
Embedding
System
Proposed Chaotic Key Generator
Stego Image
Payload Redundancy Removal
45
4.3.2 PROPOSED SYSTEM IMPLEMENTATION
The system is explained in three stages
A. Finding The Adaptive Chaotic Key
B. Proposed 13-bit Representation
C. Embedding System
The three stages are explained in detail in the subsequent parts of this section.
A. Finding the Adaptive Chaotic Key
We have proposed the method for key generation as follows:
The key generation has been made dependent on the result of chaotic key generator (CKG),
which generates an output depending on these conditions:
a) If 0 ≤ rand < 0.25 output is Z1
b) If 0.25 ≤ rand < 0.5 output is Z2
c) If 0.5 ≤ rand < 0.75 output is Z3
d) If 0.75 ≤ rand < 1 output is Z4
Where Zi are integers ( i = 1, 2, 3, 4)
Suppose the sequence of output integers from CKG is [Z1, Z2, Z3….]
Location of first pixel = Z1
Location of second pixel = (Z1 + Z2)
Location of third pixel = (Z1 + Z2 +Z3) and so on.
Fig 4.11: Chaotic Key Generator
Chaotic Key Generator
Current Location
Z1
Z2
Pixel location
46
The average separation this algorithm would provide between the pixels to embed information is
dependent on the value of possible output of chaotic key generator that is given as:
Mean separation between pixels = ∑ R S�� / 4
.The average separation of the key generator is dependent on size of information to be embedded
and size of cover image. The following algorithm explains stepwise the calculation of average
separation of the chaotic key. Figure 4.12 given below follows these steps.
Fig 4.12: Adaptive Key Generation
Steps:
1. Two dimensional DCT is applied on the M x N message image
2. Resultant DCT equivalent is passed through thresholding block to eliminate the small
magnitude DCT coefficients which are not required for reconstruction. Reduced
dimensions of the image after thresholding are calculated by locating the row and column
index containing the last non zero entry. These are named as M’ x N’.
3. Total number of bits in the reduced block of information is calculated by using relation :
B = M’ x N’ x 8; considering 8-bit format.
Message Image
(MxN) 2-D DCT Thresholding
Reduced Dimensions
(M’x N’)
Total bits B=M’xN’x8
(n,k) channel coding
Bc=(B/k)*n
Pixels required to embed 3
bits/pixels R=Bc/3
Average Separation Required to
embed S=P/R
Cover Image AxB
Total Pixels P=AxB
Adjust Key
Redundancy Removal Block
47
4. Information bits are passed through (n,k) Reed Solomon encoder. We have used (15,7) 2-
bit error correction code. So total number of bits becomes Bc= B.n/k
5. Number of pixels in cover image is P = A x B; where A and B are dimensions of cover
image
6. Since 3 bits are being embedded per randomly chosen pixel , So number of pixels
required for embedding all bits is given as: R = Bc/3
7. The average separation required between the pixels of cover image to embed information
is calculated as S= P/R
8. The selected average is used to find the values zi, (i=1, 2, 3, 4) to be used for key
generation
Following small example illustrates the process. If S = 5, we could have
Z1 = 2, Z2 = 4, Z3 = 6 and Z4 = 8
Such that ∑ R S�� / 4 = 5
Similarly we could have Z1 = 1, Z2 = 4, Z3 = 6 and Z4 = 9 Such that ∑ R S�� / 4 = 5.
We can choose any sets of �R���S
B. Proposed 13-bit Representation System
As we discussed the focus of our research is to enhance the available region for embedding
information. We present a new representation of cover image to enhance the capacity to embed
more information. If we consider the normal bit representation of a pixel value it lies in the range
0-255 we use 8-bit binary representation. For example the value 51 in decimal is represented in
8-bit format as 00110011.
In LSB embedding schemes the least significant bit is used for embedding that has weight =1,
which resultantly doesn’t affect the pixel value much that is maximum by addition or subtraction
by a value 1. So we need a representation system which provides more least significant levels to
embed more information.
For the above mentioned reason we have proposed a new representation system that is based on
prime number series, for which we have extended the 8-bit representation to 13 bit representation
in order to add more bit planes that can be used to embed more information. In this case the basis
48
is a set of prime numbers from 1 to 41 that are: (1, 3. 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41). The
scheme is explained to represent 51 in terms of these new basis as below.
Bit
Weight
1 3 5 7 11 13 17 19 23 29 31 37 41
Bit
Value
1 0 0 0 0 1 0 0 0 0 0 1 0
Decimal value is represented in these new prime number basis as given above so 51 = 1 + 13 +
37. We can represent any gray scale pixel value between 0 - 255 using this representation. Thus
the pixel values of cover image if represented using this proposed representation have more bit
plans to offer for embedding which is basically the aim of this research to improve capacity.
Furthermore it is difficult for an attacker to judge which representation has been used for
embedding. So as compared to simple LSB embedding this scheme works better in terms of
capacity and security as well. We have used last 3 significant bits for data embedding which
increases the capacity of the proposed embedding scheme three times as compared to LSB
embedding. To reduce the computational load we have converted only those pixels of cover
image that have been randomly chosen for embedding information.
C. Embedding System
The overall implementation of the proposed embedding system is presented in Figure 4.13.
Fig 4.13: Embedding System
Bc Message bits Calculated in
Algo
Cover Image
Optimized Key From Algo
Random Pixel Selection
Conversion of selected pixels to
13 bit prime series Representation
Embedding 3 bits per randomly chosen pixel
Stego
49
The inputs to this module are the key from the chaotic key generator, the encoded information
bits and the cover image. Following steps show the embedding:
Steps:
1. Generated key is used to locate the pixels to be used for embedding
2. The chaotically located pixels are converted into 13-bit format using prime number
representation
3. 3-bits of encoded information is embedded in last 3 bits of the each chaotically selected
pixel which is first converted to 13-bit system
4. The converted pixels are back converted to 8-bit format to generate stego image
4.3.3 Extraction and Recovery of Message
The algorithm for extraction of the information is shown in Figure 4.14.
Fig 4.14: Extraction and Message Recovery
The algorithm uses the chaotic key generated at encoder and stego image as inputs. The
algorithm works following below mentioned steps:
Steps:
1. The chaotic key is used to locate the pixels of the cover image that are carrying
information
2. The selected pixels are converted to 13-bit format and information is extracted from the
last tree bits of each selected pixel
3. The extracted information is channel coded, So RS decoder is applied to decode the
information
Stego image
Pixel selector
13-bit conversion And information
extraction RS decoder Inverse DCT Message
Key
50
4. The decoded information is passed through inverse DCT transform to recover the
message image
4.3.4 SIMULATION RESULTS
We would be presenting results obtained using the proposed algorithm in this section. The two
images of size 256 x 256 shown in Figure 4.15 are selected as message images for embedding.
The two different images that are being used as cover images are shown in Figure 4.16, both of
size 512 x 512. As we discussed we have applied DCT in combination with thresholding to the
message image in order to remove the redundancy in payload. We have used 2-bit error
correction Reed Solomon code to make the message recovery error free. Table 4.3 demonstrates
the significant payload reduction for message image Eagle due to DCT and thresholding being
applied on message image at different values of thresholding parameter. Results shown emphasis
that the payload is much lesser than the original message even after adding redundancy for the
correction code. The mentioned reduced block of DCT coefficients is selected for embedding so
we only need to know the size of this block at receiver side instead of the knowing location of
each non zero coefficient. The value of average separation required between the pixels of cover
image to embed information is also shown against each value of thresholding, that has been
calculated by our proposed algorithm and the chaotic key has been adjusted accordingly. Table
4.4 demonstrates the similar results for message image 2.
Fig4.15: (a) Message Image Eagle (b) Message Image Roger
51
Fig 4.16: Cover Images (a) Baboon (b) Lena
Table 4.3: Payload and Average Separation Calculation for Message Image Eagle
Threshold
Value
(Th)
Reduced
Image
Dimension
M’ x N’
Total bits to be
Embedded
Bc
Average
Seperation
1 160 x 167 213760 2
1.5 71 x 74 90068 9
2 59 x 70 70800 12
2.5 57 x 63 61560 13
Table 4.4: Payload and Average Separation Calculation for Message Image Roger
Threshold
Value
(Th)
Reduced
Image
Dimensions
M’ x N’
Total bits to be
Embedded
Bc
Average
Seperation
1 56 x 74 71040 12
1.5 53 x 60 54515 15
2 39 x 60 40115 20
2.5 38 x 50 32572 25
52
We have selected two different performance parameters to demonstrate the results that are:
1. Peak Signal to Noise Ratio (PSNR)
2. Structural Similarity Index (SSIM)
1. Peak Signal to Noise Ratio (PSNR):
PSNR is most the widely utilized figure of merit to indicate the quality of a stego image [119].
The original cover is termed as the signal of interest and the modifications caused by embedding
information are termed as noise. PSNR is defined as:
2. Mean Structural Similarity Index (MSSIM):
The MSSIM index is a relationship developed to measure the similarity between two images.
SSIM has been designed to aid the image processing to determine the image quality which is
conventionally measured by MSE or PSNR, which are considered to be old fashioned now [120].
The MSSIM metric is calculated on various windows of an image. The measure between two
windows x and y of common size N ×M is computed as:
Where X and Y are in our case would be considered as cover and stego images respectively, xj
and yj are the image contents at the jth local window, and M is the number of windows of the
image. SSIM is computed as,
where σx is the mean intensity of x, σy is the mean intensity of y, σ2x is the variance of x, σ2
y is the
variance of y, σxy is the variance of x and y, C1 = (K1L)2, C2 = (K2L)2 are the two variables to
stabilize the division with weak denominator, L is the dynamic range of the pixel values (255 for
8-bit grayscale image), K1=0.01 and K2=0.03 by default. The value of MSSIM should be closer
(4.7)
(4.8)
(4.9)
53
to 1 to indicate the maximum similarity between cover and stego image. This maximum
similarity indicator makes it more difficult to predict a clue of data embedding.
Table 4.5 demonstrates the values of the two performance parameters mentioned above for both
the cover images when the message image embedded is Eagle. Results are presented against the
different values of thresholding applied on message image Eagle. In addition to it we have also
presented the PSNR of the recovered message image against each value of the thresholding
parameter. The values of PSNR and MSSIM found clearly depict that our proposed algorithm
results in a high capacity and the imperceptibility of the cover image is also high. In addition to it
the recovered message image is also shown in Figure 4.17 for each value of thresholding
parameter which was embedded after thresholding resulting into improved capacity due to
payload reduction.
Fig 4.17: Recovered Message Image Eagle for Different Values of Thresholding Parameter (a)
Th = 1 (b) Th = 1.5 (c) Th = 2 (d) Th = 2.5
Table 4.5 presents the similar results for message image Roger as presented in Table 4.6. Figure
4.18 shows the recovered message image Roger for the different values of thresholding being
applied.
54
Fig 4.18: Recovered Message Image Roger for Different Values of Thresholding Parameter (a)
Th =1 (b) Th = 1.5 (c) Th = 2 (d) Th = 2.5
Table 4.5: Simulation Results for Message Image Eagle
Threshold
Parameter
(Th)
PSNR of
Baboon
(dB)
PSNR of
Lena (dB)
MSSIM of
Baboon
MSSIM of
Lena
PSNR of
Recovered
Image (dB)
1 60.34 60.13 0.9877 0.9859 58
1.5 61.78 61.22 0.9921 0.9908 57.2
2 63.05 62.96 0.9958 0.9940 56.8
2.5 63.78 63.41 0.9970 0.9961 56.56
Table 4.6: Simulation Results for Message Image Roger
Threshold
Parameter
Th
PSNR of
Baboon
(dB)
PSNR of
Lena (dB)
MSSIM of
Baboon
MSSIM of
Lena
PSNR of
Recovered
Image (dB)
1 61.45 60.96 0.9884 0.9876 59.61
1.5 62.98 62.13 0.9945 0.9923 58.92
2 63.66 62.98 0.9978 0.9965 58.24
2.5 63.97 63.27 0.9991 0.9987 57.83
55
Results presented in Table 3 and 4 along with recovered images shown in Fig.9 and 10 clearly
portray that as we increase the value of thresholding parameter the size of payload reduces which
results in improvement in the PSNR and MSSIM of stego image. But the PSNR of recovered
image falls off because of more information content being reduced. Still the recovered image
with lowest PSNR is also clearly readable which emphasizes on reliability of the proposed
system.
4.4 SUMMARY
The first proposed work primarily exploits image steganography with a blend of right translated
gray S-box and chaotic differential LSB embedding. The chaotic embedding is further aided by
genetic algorithm (GA) to minimize the distortion in cover image by shuffling the positions of
the pixels to embed the information and selecting the optimized locations yielding higher values
of PSNR and minimum MSE. The super imposition of BCH codes is an additional step towards
security improvement making the secret information barely visible and hard to interpret.
Simulation results illustrate high PSNR and low MSE value depicting high imperceptibility
achieved by the proposed algorithm. To explore transform domain for the described embedding
and encryption techniques to improve the robustness of the stego image is our future work.
In the second proposed system we explored a new representation of cover image pixels to exploit
3 bit planes for embedding the information to improve capacity. The improvement in capacity is
further aided by the reduction in the payload done by using DCT and thresholding on the
message image. The message image has been given a cover of 2-bit error code to make it robust
against errors. The system is able to automatically adjust the chaotic key depending on the cover
image and the message image. The results presented demonstrate better capacity and
imperceptibility using two different performance evaluation parameters. The message image
recovered is of good quality and is clearly readable depicting error free recovery.
56
Chapter 5
COMPRESSED SENSING BASED
IMPROVEMENT IN CAPACITY AND
SECURITY OF IMAGE
STEGANOGRAPHY FOR SECRET
IMAGE AND SECRET AUDIO MESSAGE
5.1 INTRODUCTION
The proposed Image steganography system mainly focuses on enhancing the secrecy and
security of the secret message. We have proposed an innovative scheme which is based on
compressive sensing methodology for images. Compressive sensing is used in various fields for
reconstruction of information with very few measurements. The measurements are basically the
projection of original information onto the measurement basis, which are of less dimension as
compared to the original information. The measurement basis are only known to the intended
user. Without the knowledge of measurement basis, reconstruction of information is impossible.
This feature of compressive sensing is a captivating factor to employ it for security enhancement
of the information. The additional benefit achieved is the payload reduction as we are able to
reconstruct the original information from very less measurements.
This chapter is based on two of our research contributions based on compresses sensing. The first
contribution deals with embedding secret image within an image and the second one deal with
embedding an audio in the cover image. The second contribution is based on the proposed work
presented in first contribution. Considering the first contribution the secret image is sparsified in
DCT domain by using thresholding. The sparsified signal is then impressed upon the
measurement basis matrix which not only enhances the security but also compresses further the
information to be embedded in a cover image. The embedding is done using conventional LSB
technique. However the pixels are chosen using the numbers generated by the chaotic equation.
57
At receiver side we have reconstructed the grayscale image successfully using compressive
sensing recovery algorithm.
In second contribution the secret message is an audio clip. We already had designed an efficient
steganography system based on compressed sensing for embedding images, so we used it for
audio clip instead of designing a new system. To employ the proposed framework for images in
the case of audio clip firstly, we have converted the audio message to an equivalent grayscale
image. This image has been used as secret image for the proposed image steganography system
explained earlier. All the remaining steps are same as in the first contribution. The audio
message is regenerated using the reconstructed information. In the nutshell compressive sensing
has resulted in huge secrecy enhancement, as well as, increase in payload capacity in the cover
image.
The results presented show that the proposed system was able to successfully reconstruct both
image and audio message at the receiver side with a good PSNR value even after various image
processing attacks being applied to the system. The presented results emphasize that the
proposed model is highly imperceptible while at the same time offer robustness against various
attacks.
The rest of the chapter is organized as follows. First section explains the image steganography
system based on compressed sensing for secretly transmitting images. The second section
discusses the utilization of the proposed system explained in section I for transmitting an audio
message.
58
5.2 COMPRESSED SENSING BASED IMPROVED
IMPERCEPTIBILITY AND SECURITY IN IMAGE
STEGANOGRAPHY SYSTEM
5.2.1 PROPOSED MODEL
Block diagram of the proposed model is presented in figure 1. The process starts by applying
compressive sensing to the secret image containing the message which in a single step
compresses and encrypts the secret image, following to that the compressed information is
embedded to the cover image. This idea can be explained in three modules that are:
A. Compressive sensing
B. Embedding Procedure
C. Message Reconstruction and Image Recovery:
A. Compressive Sensing:
Compressive sensing is widely used to reconstruct signals with much reduced set of information.
We have used this attribute of compressive sensing to improve the steganography system in
terms of capacity, imperceptibility and security altogether. The block diagram explaining each
step of compression and reconstruction is shown in figure 3. Secret image containing message is
termed “f” is the input for this module. The steps involved in this process can be explained
sequentially as follows:
1. The 2D DCT transform is applied to the “nxn” input image containing secret
message using the relation:
X=H*f*HT
2. The “nxn” matrix “X” containing DCT coefficients is passed through thresh
holding module which only retains significant coefficients. All remaining
coefficients are set to zero
3. The sparsified matrix Xs is than read column by column and each column is
projected to the measurement matrix individually using the relation:
Y(i)=ΦUVP(i)
59
This module further reduces the dimensions of the image as the measurement matrix “Φ” is of
dimension “mxn” where m is much lesser than n.
4. After n iterations of step 3 the resultant columns are concatenated to form matrix
Y. this Y matrix is of size “mxn” which is fed to embedding module.
Fig 5.1: Encoder
B. Embedding:
The binary information which is the output of the module 2 corresponding to each DCT
coefficient is than embedded to LSBs of chaotically selected pixels of the cover image in a
similar way as we have explained in section 3.4.3 and section 4.2.
C. Message Reconstruction and Image Recovery:
At the decoding side the reverse process is applied as shown in figure 2. The chaotic key is
available at decoder so same chaotic sequence is generated to locate the pixels of stego object
containing hidden information. Figure 3 shows the reconstruction process after de embedding.
The input to reconstruction is the de-embedded information which is matrix Y. Now in
reconstruction of secret image steps involved are:
a) Pick one column of recovered Y and apply l1 minimization subject to constraint as
given below:
2D- DCT X=H*f*HT
Secret Image (Message “f”)
XS : Sparsified Image after thresholding
For i=1:n Pick UVP(i) ith column
of XS
Y(i)=ΦUVP(i) Embedding into Cover Image
i ≤ n
|X|≤Xth Thresholding
Stego Image
Yes
No
60
min WUXP���W1 Y. Z �[�\� ) ]_̂`�\��aa b c
Where c is measurement noise variance.
This can also be written in unconstrained lagrangian form as:
minWUXdW� + ,W*��� ) �UXP���W((
Here UXP��� is the estimate of ith column of sparsified DCT coefficient matrix of secret image
found by solving this optimization problem
b) After repeating the step 1 “n” times the resultant columns of each iteration are
concatenated to find estimate of sparsified DCT coefficient matrix UXP of secret
image .
c) The 2D inverse DCT transformation is applied to the estimated UXP to get the
image back using following relationship:
�e=HT*UXP *H
Fig 5.2: Decoder
Stego Image De-embedding
for i=1:n, Pick ith column from de-embedded
information and apply CS recovery algorithm
i ≤ n
Formulate UXP matrix from
UXP���
Inverse2D-DCT �e=HT*UXP *H PSNR
Yes
No
61
5.2.2 Advantages of using Compressed Sensing in Image Steganography
Compressed sensing is based on two different types of basis which have been explained in detail
in section 3.2. The advantage we have gained by incorporating compressed sensing is due to the
measurement basis Φ. The main advantages gained are listed below:
1. Remarkable increase in the security of the secret information
2. Significant reduction in the payload
Now we explain the details that how measurement basis helped for achieving these above
mentioned advantages which are considered to be prime objectives of a steganography
algorithm. The key factor involved to attain these advantages is the generation of the
measurement basis. For instance in our case the basis are generated by generating randomly mxn
matrix which contains entries either +1 or -1 as its enteries.
This means we can mathematically say the elements of Φmxn are Φij Є {1,-1}. This shows that
each element of the measurement basis can either be 1 or -1. So each element can have two
possible values which make it similar as a binary case. Now if we consider the total possible
combinations of the entries of matrix Φmxn they would be:
Possible combinations of the entries = 2mxn
This is because each element can have two values either 1 or -1 and the total number of elements
in Φ are (m x n).
Consider an example. Let Φ be of size 140 x 512. Where m = 140 and n = 512. The total possible
combinations against this case would be
Possible Combinations = 2140 x 512 = 271680 = 6.76 x 1021577 combinations.
It is clear from the above example that it is computationally impossible to generate the basis by
any attacker because the generation procedure is also unknown as there is no information about
the entries of the basis and secondly, the total possible combinations is too huge. So it is clear
from the above discussion that the measurement basis enhance the security by a huge level.
62
The payload reduction is a key feature of compressed sensing as it works on reconstructing
information with few available measurements. This attribute of CS framework helped us to
reduce the payload significantly, while transmitting the same amount of information which
helped in increasing the payload capacity and imperceptibility of the system. The calculation of
payload reduction and comparison with the scheme presented in section 4.3 which did not utilize
CS has been shown in Table 5.3.
5.2.3 SIMULATION RESULTS AND DISCUSSION
This section presents the simulation results using the proposed steganography system tested at
different conditions for multiple secret images. Simulation results presented show that there is a
significant reduction in the Payload embedded to the cover image and secrecy is also improved
as a result of using compressed sensing. Results clearly demonstrate that the secret image has
been reconstructed successfully by transmitting less information and the message is still clearly
readable. The cover image used is lena of size 512x512 shown in Figure 5.3(a). Two different
secret images of size 384x384 are used as secret message. The measurement matrix used is
randomly generated containing values +1 and -1 which is only known to intended user making
system robust to attacks and secrecy is enhanced. Figure 5.3(b) shows the case of secret image
containing hand written text message. Figure 5.3(c) shows the case when the secret image
contains typed text message. Figure 5.4 shows the case related to hand written message for
different values of “m” which determines the number of rows of measurement matrix. We can
see that for smaller values of “m” the reconstructed image is blurred. We have tried to show the
range of “m” which covers the minimum clear reconstruction to fine quality reconstruction of
secret image. Figure 5.5 demonstrates the same results for secret image containing written text.
63
.
Fig 5.3(a) Fig 5.3(b) Fig.3(c)
64
Fig 5.4(a-e): Reconstructed Secret Image with m=40, 80, 120, 160 and 200
The results presented in figure 4(a-e) reflect that with increase in the dimension of the
measurement matrix represented by the parameter “m” improves the quality of reconstructed
hand written secret image. The tabular form of these results indicating the value of “m” and
corresponding PSNR of reconstructed secret image is given in Table 5.1. The threshold value for
sparsifying the image as shown in block diagram in figure 2 is used as 2.5 means all the DCT
65
coefficient lesser than this value are set to zero. So only 993 coefficients out of 147456 are left
reducing the payload significantly.
Table 5.1: PSNR of Reconstructed Handwritten Secret Image
Number of rows of measurement matrix (m)
PSNR of Stego Image PSNR of reconstructed secret Image
40 63.12dB 53.74dB
80 62.93dB 54.91dB
120 62.69dB 56.13dB
160 62.48dB 57.69dB
200 62.36dB 58.93dB
66
Fig 5.5(a-e): Reconstructed Secret Image with m=40, 80, 120, 160 and 200
Figure 5.5(a-e) presents similar results presented in previous figure but for secret image
containing typed text secret message, we can clearly see the quality of image is improving as we
increase parameter “m”. but the noticeable thing is the reconstructed secret image is readable
with very less measurements which is the desired result of the proposed system. Table 5.2
presents the value of PSNR of reconstructed image for different dimensions of measurement
matrix. The threshold for this case is set to 0.8 so 1329 coefficients are left out of 147456
coefficients. Another noticeable result indicated in the table 1 and table 2 is the PSNR of stego
image which is good in all cases depicting the essence of using compressive sensing as the size
of payload is very much reduced by the proposed algorithm.
Table 5.2: PSNR of Reconstructed Typed Secret Image
Number of rows of measurement matrix (m)
PSNR of Stego Image PSNR of reconstructed secret Image
40 62.76dB 60.78dB
80 62.47dB 63.87dB
120 62.28dB 66.37dB
160 62.02dB 66.58dB
200 61.88dB 66.93dB
67
The results presented in Table 5.1 and 5.2 clearly show that increase in the size of m helps
reconstruction of secret image at higher PSNR but slightly effecting the PSNR of stego image.
We can improve the PSNR of the reconstructed image by slightly increasing the threshold we
applied at sparsifying stage but this would result in increasing the payload size which in return
degrades the stego image quality.
To compare the payload reduction achieved in this case for the message image eagle and the
payload reduction achieved for the same message image in the scheme presented in section 4.3
we have presented Table 5.3.
Table 5.3: Comparison of the proposed scheme with the scheme presented in Section 4.3 in
terms of payload reduction
Threshold value 1 1.5 2 2.5
Payload 160 x 167 71 x 74 59 x 70 57 x 63
No of Rows in φ 160 120 80 40
Payload 160 x 384 120 x 384 80 x 384 40 x 384
The top two rows correspond to the scheme presented in section 4.3. In that case payload is
dependent on the threshold parameter; we can see that the payload reduces significantly when
thresholding parameter is increased. The bottom two rows show the payload reduction achieved
in CS based technique. Here the payload size is dependent on the value of “m” which is the
number of rows in measurement basis, we can see payload is minimum when maximum
compression is applied. If we compare the two techniques on the basis of payload reduction the
scheme presented in section 4,3 performs better in comparison with the technique based on CS,
but CS based technique outperforms in terms of secrecy as discussed in detail in section 5.2.2
while it also reduces the payload.
68
5.3 COMPRESSED SENSING BASED IMAGE
STEGANOGRAPHY SYSTEM FOR SECURE TRANSMISSION
OF AUDIO MESSAGE WITH ENHANCED SECURITY AND
CAPACITY
We propose an innovative information security system which utilizes compressed sensing in
steganography for secretly transmitting an audio message. The proposed system converts an
audio message to a grayscale image that is sparsified using 2D-DCT alongwith thresholding.
Compressed sensing further compresses and also adds security to the message. The processed
secret message is embedded in the cover image using conventional LSB technique and
chaotically selected pixels in the cover image.
5.3.1 TRANSMITTER SIDE
The block diagram of the proposed transmitter side that comprises of secrecy enhancement and
embedding of the secret message is shown in Figure 5.6.
Fig 5.6: Secrecy Enhancing and Embedding of the Secret Message Using Compressed Sensing
Yes
Compressed Sensing Block
Secret Audio
Message
Sampling and Quantization of samples to 8-bit
Array
Reshape array of samples to square matrix by padding
zeros (Grayscale Image)
2D DCT Thresholding |X|≥Xth
Sparse Image Xs
For i= 1: n Pick ith
column of Xs i.e. UVP(i)
Compute Y(i)=ΦUVP(i)
i≤n Formulate Matrix Y by combining Y(i) for i= 1:n
Embed
Cover Image
Stego Image
No
69
The steps involve in the process are:
Steps:
1. Reading Audio Message:
The continuous audio message is sampled and quantized to 8-bit format and is saved as an array
of samples
2. Reshaping Array of Samples to Square Matrix:
The array of samples is transformed into the closest square matrix orientation where the
additional values are filled by using zero padding. The resultant “nxn” square matrix is a
grayscale image corresponding to the secret audio message
3. Applying 2-D Discrete Cosine Transform:
The 2D DCT transform is applied to the “nxn” input grayscale image containing secret message
using the relation:
X=H*f*HT
4. Thresholding of the Coefficients:
The “nxn” matrix “X” containing DCT coefficients corresponding to the grayscale image is
passed through thresh holding module which only retains significant coefficients. All remaining
coefficients are set to zero
5. Applying Measurement basis:
The sparsified matrix Xs is than read column by column and each column of size “n x 1” is
projected to the measurement matrix “Φ” of dimension “mxn” that contains randomly generated
±1 and is only known to intended user individually using the relation:
Y(i)=ΦXS(i)
The resultant column Y(i) is of the dimension “m x 1” which is much lesser than Xs(i) because
m<<n.
70
6. Reshaping Information for Embedding:
After n iterations of step 5 the resultant columns are concatenated to form matrix Y. this Y
matrix is of size “mxn” which is ready to be embedded to the cover
7. Embedding the Secret Compressed Information in the Cover Image:
Procedure for embedding is explained in following steps:
I. The pixels of the cover image for embedding are randomly selected using chaotic
key
II. Each selected pixel is converted to binary 8-bit format
III. One bit of information is embedded by using LSB replacement of the cover image
pixel by the information bit
IV. The pixel is converted back to intensity level
V. After comlete information is embedded the stego Image is ready to be transmitted
5.3.2 DECODING AND RECONSTRUCTION OF SECRET MESSAGE
The decoding and recovery of the secret message is illustrated in Figure 5.7.
Fig 5.7: De-Embedding and Decoding of Image using CS Recovery Algorithm
No
Inverse DCT
Recovered Grayscale
Image
Stego Image
De-Embed and Extract
Y
Pick ith Column of Y and Apply Recovery
Algorithm i ≤n
Formulate matrix UXP by combining
UXP��� for i= 1:n
Matrix to Array
Conversion Recovered
Audio
Yes
71
The steps involved in the reconstruction of the secret message are:
Steps:
The same chaotic key used at transmitter side is used to locate the pixels containg information in
the stego image. The information is extracted and given as the form of “m x n” matrix Y.
1. Applying l1 Minimization to Reconstruct Information:
Pick one column of recovered Y and apply l1 minimization subject to constraint as given below:
min WUXP���W1
Y. Z �[�\� ) ]_̂`�\��aa b c
Here UXP��� is the estimate of ith column of sparsified DCT coefficient matrix of secret image
found by solving this optimization problem and c is noise variance.
2. Reshaping the Reconstructed Information to Matrix:
After repeating the step 2 “n” times the resultant columns of each iteration are concatenated to
find estimate of sparsified DCT coefficient matrix UXP of secret image.
3. Applying DCT Inverse:
The 2D inverse DCT transformation is applied to the estimated UXP to get the image back using
following relationship:
�e=HT*UXP *H
4. Conversion to Audio:
The reconstructed image is contains the samples of audio message which are again given the
form of an array to play the message
72
5.3.3 SIMULATION RESULTS
This section presents the simulation results generated using the proposed algorithm. The selected
cover image lena of size 512 x 512 is shown in Figure 5.8. The grayscale image of size 288 x
288 generated against the secret audio message of 10 seconds duration is shown in Figure 5.9.
We can see in Figure 5 the grayscale image generated against the audio message conveys no
information if recovered until someone knows the conversion that was done at transmitter.
Fig 5.8: Cover Image
Fig 5.9: Grayscale Image generated against audio message
To emphasis on the robustness of the proposed algorithm we have tested our algorithm for four
types of image processing attacks that are addition of following types of noise to the system:
1. Gaussian Noise
2. Salt & Pepper Noise
3. Speckle Noise
4. Poisson Noise
73
The algorithm is tested by varying the noise variance for the above mentioned types of noise. For
each value of noise variance the results are generated for the different values of parameter “m” of
compressed sensing that is basically the number of rows of the measurement matrix. The value
of “m” determines how much the information is being compressed i.e. lesser the value of “m”
more is the compression attained. So the payload is minimum when “m” is minimum.
Considering the stego image PSNR has been selected as a figure of merit to indicate the
imperceptibility achieved after embedding of data, while the PSNR of the regenerated audio
signal using the reconstructed grayscale image is used to show the audio message can be clearly
understood at receiver side
Table 5.4 presents the PSNR of the reconstructed audio message when the Gaussian noise is
added to the system. Considering the row wise presented results, the first row presents the case
when there is no noise added. The PSNR of audio message is presented at different values of
“m”. We can clearly see the PSNR improves when m is increased because when increasing
dimension of measurement matrix more information is retained in the compressed grayscale
image which results in better reconstruction of the audio message.
Following the first row the remaining rows present the similar results when the noise is added to
the system. Five rows of results following the first row present the PSNR values of reconstructed
audio message against noise variance ranging between 0.01 to 0.05. The presented results show
the decrease in PSNR of the reconstructed audio message as noise variance is increased but the
trend of PSNR improvement with increase in “m” is same for all the presented results.
Table 5.4: PSNR of recovered audio in dB with added Gaussian noise:
m
var
80 100 120 140 160 180
0 61.09 62.13 63.09 64.22 64.87 65.21
0.01 58.11 59.47 60.39 61.71 61.88 62.90
0.02 57.43 58.62 59.76 60.96 61.03 61.89
0.03 56.8 57.87 58.97 60.03 60.14 61.04
0.04 56.03 57.01 58.09 59.23 59.56 60.44
0.05 55.61 56.34 57.57 58.66 58.89 59.38
74
Figure 5.10 shows the graphical representation of the presented results in Table 1. We can see in
the graph presented in Figure 5.9 that the best results are achieved when m = 180.
Fig 5.10: PSNR of Recovered audio message with Gaussian Noise
Table 5.5 presents the PSNR of the stego image considering the same format used for audio
message in Table 5.4.
Table 5.5: PSNR of stego Image in dB with Gaussian Noise
m
var
80 100 120 140 160 180
0 63.19 62.34 61.93 60.77 60.11 59.88
0.01 60.17 59.39 58.78 57.51 57.25 56.90
0.02 59.56 58.13 57.66 56.66 56.37 56.04
0.03 58.84 57.67 56.97 55.86 55.59 55.49
0.04 58.03 56.88 56.13 55.03 54.96 54.34
0.05 57.51 56.19 55.48 54.39 54.09 53.68
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.0554
56
58
60
62
64
66
Noise Variance
PSNR(dB)
PSNR of Recovered Audio Message With Gaussian Noise
m=80
m=100
m=120
m=140
m=160
m=180
75
Figure 5.11 shows the stego image when the embedding is done at m = 180, the effect of noise
addition is clear as we can see increasing the value of noise variance further degrades the stego
image which in return is responsible for the degradation of PSNR of stego image and the
reconstructed audio message.
Fig 5.11: Stego Image with Gaussian Noise variance (a) Zero (b) 0.01 (c) 0.02 (d) 0.03 (e) 0.04
(f) 0.05
76
Table 5.6 and Table 5.7 present the similar results presented in Table 5.4 and Table 5.5 for salt &
pepper noise.
Table 5.6: PSNR of audio message in dB with Salt and Pepper noise
m
var
80 100 120 140 160 180
0 61.09 62.13 63.09 64.22 64.87 65.21
0.01 59.23 60.03 60.98 62.12 62.77 63.03
0.02 58.45 59.23 60.10 61.57 61.89 62.23
0.03 57.88 58.51 59.23 59.88 61 61.45
0.04 57.02 57.87 58.49 58.93 59.21 59.63
0.05 56.49 57.10 57.78 58.01 58.45 58.76
Figure 5.12 shows the graphical representation of results presented in Table 3. This graph also
follows the same trend as shown in Gaussian case. The PSNR drops with increase in noise
variance for fixed “m”. The best resuts are attained when “m” is maximum.
Fig 5.12: PSNR of audio message recovered with salt & pepper noise added
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.0556
57
58
59
60
61
62
63
64
65
66
Noise Variance
PSNR(dB)
PSNR of Recovered Audio Message With Salt and Pepper Noise
m=80
m=100
m=120
m=140
m=160
m=180
77
Table 5.7: PSNR of stego image in dB with Salt and Pepper noise
M
Var
80 100 120 140 160 180
0 63.19 62.34 61.93 60.77 60.11 59.88
0.01 61.28 60.47 59.79 58.61 57.97 57.30
0.02 60.47 59.80 58.98 57.88 57.04 56.74
0.03 59.79 58.97 58.07 57.02 56.50 55.89
0.04 58.97 58.18 57.48 56.43 55.93 55.01
0.05 58.11 57.69 56.76 55.68 55.18 54.37
The stego image for m=180 for the different values of noise variance is shown in Figure 5.13.
78
Fig 5.13: Stego Image with Salt & Pepper Noise variance (a) Zero (b) 0.01 (c) 0.02 (d) 0.03 (e)
0.04 (f) 0.05
Table 5.8 and Table 5.9 present the results when speckle noise is added to the system. Results are
presented in the similar way as in Table 5.4 and Table 5.5.
Table 5.8: PSNR of audio message in dB with Speckle noise
m
Var
80 100 120 140 160 180
0 61.09 62.13 63.09 64.22 64.87 65.21
0.01 59.50 60.19 61.30 62.43 62.91 63.4
0.02 58.89 59.41 60.61 61.68 62.01 62.65
0.03 58.03 58.79 59.88 60.77 61.27 61.85
0.04 57.46 57.97 58 59.83 60.61 61
0.05 56.78 57.28 57.56 58.91 59.71 60.13
Figure 5.14 shows the results presented in Table 5.7 in graphical form.
79
Fig 5.14: PSNR of audio message with speckle noise
Table 5.9: PSNR of Stego Image in dB with speckle noise
M
var
80 100 120 140 160 180
0 63.19 62.34 61.93 60.77 60.11 59.88
0.01 61.46 60.67 59.96 58.83 58.17 57.49
0.02 60.78 59.96 59.18 58.01 57.54 56.92
0.03 59.98 59.04 58.41 57.32 56.82 56
0.04 59.09 58.38 57.63 56.56 56.03 55.41
0.05 58.11 57.76 56.98 55.89 55.29 54.77
Figure 5.15 shows stego image at m=180 for different values of noise variance.
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.0556
57
58
59
60
61
62
63
64
65
66
Noise Variance
PSNR(dB)
PSNR of Recovered Audio Message With Speckle Noise
m=80
m=100
m=120
m=140
m=160
m=180
80
Fig 5.15: Stego Image with Speckle Noise variance (a) Zero (b) 0.01 (c) 0.02 (d) 0.03 (e) 0.04 (f)
0.05
Table 5.10 presents the PSNR of reconstructed audio message and Table 5.11 presents PSNR of
stego image when poison noise is added to the system. These results are different in a way that
poison noise added has a fixed variance. So we have presented results here with no noise added
and after noise addition.
81
Table 5.10: PSNR of audio message in dB with Poison noise added
M 80 100 120 140 160 180
No noise 61.09 62.13 63.09 64.22 64.87 65.21
Poison
Noise
Added
60 61.09 62.10 63.18 63.79 64.14
Graphical representation of Table 5.10 is presented in Figure 5.16. We can see the increase in
PSNR with the increase in the dimension of measurement basis.
PSNR of audio message in dB with Poison noise added
Fig 5.16: PSNR of recovered audio with Poison Noise added
Table 5.11: PSNR of Stego Image in dB with Poison noise added
m 80 100 120 140 160 180 No noise 63.19 62.34 61.93 60.77 60.11 59.88
Poison Noise Added
60.19 59.29 58.80 57.68 57.01 56.76
80 90 100 110 120 130 140 150 160 170 18061
61.5
62
62.5
63
63.5
64
64.5
65
65.5
Rows in PHI (m)
PSNR(dB)
PSNR of Recovered Audio Message With Poisson Noise
82
Figure 5.17 shows the Stego image with poison noise added.
Fig 5.17: (a) Stego Image with no noise (b) With poison noise
All the presented results demonstrate the fact that there is decrease in the PSNR of the recovered
audio message and the stego image when the noise is added to the system. The recovered audio
has maximum PSNR when parameter “m” is maximum while the PSNR of the stego image is
minimum at that time as the embedded information is maximum. The results also present the fact
that system performs well even when the noise variance is maximum as the recovered audio has
a good value of PSNR at maximum noise addition. These results emphasis on the robustness and
security of the proposed system as system performs well in the presence of attacks done by
intruder by adding different kinds of noise to the system.
5.4: SUMMARY:
In this chapter we have explained the application of compressed sensing in steganography to
enhance security, secrecy and capacity for secret image and secret audio message in cover image
same framework has been used for both types of secret messages. For the case of secret image
the the input to the system is directly the secret image, and for the secret audio message the input
the audio is converted to a grayscale image which is then fed into the proposed system. The
grayscale secret image is sparsified by using 2D-DCT alongwith thresholding. The sparsified
image is further compressed using compressed sensing algorithm, resulting into improved
83
security and payload capacity. The compressed information is embedded in chaotically chosen
pixels of cover image using LSB embedding. The information is reconstructed at the receiver
side using compressed sensing reconstruction algorithm based on l1 minimization. The results
presented highlight the fact that the system can reconstruct the secret image with high PSNR and
also clearly audible audio message with high PSNR has been successfully recovered at the
receiver side and the system performs well in the presence of various image processing attacks.
For further improvement and enhancement in the proposed model in future, we are interested in
exploring different methodologies other than l1 minimization for reconstruction of the
information.
84
Chapter 6
CONCLUSIONS AND
FUTURE WORK
6.1 CONCLUSIONS
In this thesis we have mainly focused on improving capacity, imperceptibility and robustness of
the image steganography system. We have explored different innovative methodologies to
integrate with the image steganography for the performance enhancement of the system.
To enhance the security of the secret text information we have focused on pre-processing of the
secret text information. For this purpose right translated gray S-boxes were utilized to improve
the security of the secret information. The utilized right translated gray S-boxes is a recently
devised method in the literature which performs better than the conventional advanced
encryption standard (AES). For error free recovery the information is given a cover of BCH error
correction codes. The incorporation of error correction codes makes it possible to correct two
corrupted bits which resultantly improves the reliable and correct recovery of the information.
For the improvement of the imperceptibility of the stego image the selection of cover image
pixels used for embedding information is optimized using Genetic Algorithm (GA) keeping
PSNR of the stego image as fitness function. The information is embedded in the selected pixels
using differential LSB embedding which performs far better than the conventional LSB
embedding in terms of avoiding propagation of error and performs at the same level in terms of
imperceptibility.
For improving the capacity of the image steganography system by three times we have proposed
a novel embedding method based on novel 13-bit prime series representation of the cover image
pixels which help to embed 3 bits per pixel of the cover image which is a huge improvement in
the capacity. The proposed work also presents an algorithm which adjusts the generation of
chaotic key on the basis of size of cover image and the size of secret information to be
85
embedded. The proposed algorithm ensures that the information has been spread uniformly and
completely into the cover image. To further improve the capacity the secret image is spasrified
and thresholded in DCT domain and a reduced block of coefficients is only embedded which
reduces the payload adding further towards improved capacity and imperceptibility.
The major contribution of this thesis is combination of image steganography and compressed
sensing. The blend of compressed sensing with image steganography provides huge improvent in
security and capacity of the system at the same time. Compressed sensing is based on
measurement basis which are generated in a random fashion. These measurement basis act as a
secret key for the system. The knowledge of measurement basis at the receiver side is a key
ingredient for reconstruction of the information and without this information the reconstruction is
impossible. But it is impossible for the attacker to predict or generate these basis because of the
dimensionality of basis and the huge possible combinations for randomly generating them. Thus
this attribute of compressed sensing makes the system highly secure making decoding of the
secret information unattainable. Due to reduced measurements the payload has been reduced
which in return increases the imperceptibility of the system. The first proposed work applies
compressed sensing on the secret image and the compressed information is than embedded to the
chaotically selected pixels of cover image using LSB embedding. It has been shown that even
with the reduced payload the secret information has been successfully reconstructed with high
PSNR.
The second contribution uses the same infrastructure used in above mentioned work but in this
case the secret message is an audio. So to utilize the proposed work for images the audio
message is first converted to an image and then the converted image is compressed using
compressed sensing and the information is embedded to the chaotically selected pixels of cover
image using LSB embedding. This contribution is a smart utilization of the framework build up
for images for an audio message. The proposed work was able to regenerate both the secret
image and secret audio message with high PSNR while maintaining good imperceptibility of the
system and compressed sensing made the system highly secure.
86
6.2 FUTURE WORK
• We have embedded the information in spatial domain in all the presented techniques.
This work can be explored in frequency domain using DCT or Wavelet Transform. The
information can be embedded in the coefficients calculated against the cover image
instead of embedding in the spatial domain. This can result in better robustness of the
system against attacks.
• We can also explore different cryptography methods to see the best option available for
better robustness of the system.
• The effect of applying 13-bit prime series representation on the cover image can be
analyzed in terms of statistical properties and histogram of the stego image, to check the
proposed system performance against different steganalysis techniques.
• For reconstruction of the information in compressed sensing based techniques the work
can be extended by using different reconstruction techniques and different methodologies
can be compared on the basis of the quality of the recovered information and also the rate
of compression achieved.
87
References
[1] Mehdi Hussain, Mureed Hussain, “ A Survey of Image Steganography Techniques”.
International Journal of Advanced Science and Technology Vol.54, May 2011
[2] ZaidoonKh. AL-Ani, A.A.Zaidan, B.B.Zaidan, Hamdan.O.Alanazi, “Overview: Main
Fundamentals for Steganography”, Journal of Computing, Volume 2, Issue 3, March
2010
[3] Bin Li, Junhui He, Jiwu Huang, Yun Qing Shi, “A Survey on Image Steganography and
Steganalysis”, Journal of Information Hiding and Multimedia Signal Processing,
Ubiquitous International Volume 2, April 2011
[4] B. Li, J. He, J. Huang, and Y. Q. Shi, “A survey on image steganography and
steganalysis,” International Journal of Information Hiding and Multimedia Signal
Processing, vol. 2, no. 2, pp. 142–172, 2011.
[5] G.Swain, S.K.Lenka ”Classification of spatial domain image steganography techniques: a
study”, International Journal of Computer Science & Engineering Technology, vol.5,
no.3, pp.219-232, 2014.
[6] Fridrich, Jessica, “Steganography in Digital Media: Principles, Algorithms, and
Applications”, ISBN-0521190193, Cambridge University Press, 2010
[7] Walia, Ekta, Payal Jain, and Navdeep Navdeep. "An analysis of LSB & DCT based
steganography." Global Journal of Computer Science and Technology 10.1 (2010).
[8] Naitik P Kamdar, Dipesh G. Kamdar, Dharmesh N.Khandhar, “ Performance Evaluation
of LSB based Steganography for Optimization of SNR and MSE”. Journal of
Information, Knowledge and Research in Electronics and Communication Engineering
Volume 2
[9] ZaidoonKh. AL-Ani, A.A.Zaidan, B.B.Zaidan, Hamdan.O.Alanazi, “Overview: Main
Fundamentals for Steganography”, Journal of Computing, Volume 2, Issue 3, March
2010
88
[10] Eugene T.lin, Edward J. dell, “A Review Of Data Hiding In digital Images”,
“Video And Image Processing Laboratory School Of Electrical And Computer
Engineering
[11] Stuti Goel, Arun Rana, Manpreet Kaur, “Comparison Of Image Steganography
Techniques”, “International Journal Of Computers And Distributed Systems”, Vol .3,
Issue 1, April-May 2013. ISSN: 2278-5183
[12] K.Thangadurai and G.Sudha Devi, “An analysis of LSB Based Image
Steganography Techniques” , 2014 International Conference on Computer
Communication and Informatics (ICCCI -2014), Jan. 03 – 05, 2014, Coimbatore, INDIA
[13] Ravi Kumar, Kavita Choudhary, Nishant Dubey, “An Introduction of Image
Steganographic Techniques and Comparison”, International Journal of Electronics and
Computer Science Engineering.
[14] Mehdi Kharrazi, Husrev T. Sencar, “Performance study of common image
steganography and steganalysis techniques”, Journal of Electronic Imaging 15(4),
041104 (Oct–Dec 2006)
[15] M. Kharrazi, H. T. Sencar, and N. Memon, Image Steganography: Concepts and
Practice, Lecture Notes Series, Institute for Mathematical Sciences, National University
of Singapore, Singapore 2004
[16] Stuti Goel, Arun Rana, Manpreet Kaur, “A Review of Comparison Techniques of
Image Steganography”, IOSR Journal of Electrical and Electronics Engineering (IOSR-
JEEE) e-ISSN: 2278-1676,p-ISSN: 2320-3331, Volume 6, Issue 1 (May. - Jun. 2013), PP
41-48 www.iosrjournals.org
[17] Chen Ming, Zhang Ru, NiuXinxin, Yang Yixian, “Analysis of Current
Steganography Tools: Classifications & Features” , International Conference on
Intelligent Information Hiding and Multimedia Signal Processing (IIH-MSP'06),IEEE- 0-
7695-2745-0/06 $20.00 © 2006
[18] Dr. EktaWalia, Payal Jain, Navdeep, “An Analysis of LSB & DCT based
Steganography” ,Global Journal of Computer science & technology, Vol. 10 Issue 1 (Ver
1.0), April 2010.
89
[19] Arvind Kumar, Km. Pooja, “Steganography- A Data Hiding Technique”,
International Journal of Computer Applications (0975 – 8887), Volume 9– No.7,
November 2010.
[20] Anjali Tiwari, Seema Rani Yadav, N.K. Mittal, “A Review on Different Image
Steganography Techniques”, International Journal of Engineering and Innovative
Technology (IJEIT) Volume 3, Issue 7, January 2014
[21] Navneet Kaur, Sunny Behal, “A Survey on various types of Steganography and
Analysis of Hiding Techniques”, International Journal of Engineering Trends and
Technology (IJETT) – Volume 11 Number 8 - May 2014
[22] Artz, Donovan. "Digital steganography: hiding data within data."internet
computing, IEEE 5.3 (2001): 75-80.
[23] Amin, Muhalim Mohamed, et al. "Information hiding using steganography."
Telecommunication Technology, 2003. NCTT 2003 Proceedings. 4th National
Conference on. IEEE, 2003.
[24] Shashikala Channalli, Ajay Jadhav, “Steganography An Art of Hiding Data”
International Journal on Computer Science and Engineering Vol.1(3), 2009, 137-141
[25] Morkel, Tayana, Jan HP Eloff, and Martin S. Olivier. "An overview of image
steganography." ISSA. 2005.
[26] R. J. Anderson and Fabien A. P. Petitcolas, On the limits of steganography, IEEE
Journal on Selected Areas in Communications, vol. 16, no. 4, pp. 474-481, 1998.
[27] Niels Provos and Peter Honeyman, Hide and seek: An introduction to
steganography, IEEE Security and Privacy, vol. 1, no.3, pp. 32-44, 2003.
[28] R. Chandramouli, M. Kharrazi, and N. Memon, Image steganography and
steganalysis concepts and practice, Proc. of IWDW'03, vol. 2939, pp. 35-49, Springer,
2003.
[29] Chandra Prakash Shukla, Mr. Ramneet S Chadha, “A Survey of Steganography
Technique, Attacks and Applications”, International Journal of Advanced Research in
Computer Science and Software Engineering , Volume 4, Issue 2, February 2014
[30] A. Chedad, J. Condell, K. Curran, and P.C. Kevitt, “Digital image steganography:
Survey and analysis of current methods,” Signal Processing, vol. 90, no. 3, pp. 727–752,
March 2010.
90
[31] Chan, C. K. and Cheng, L. "Hiding data in images by simple LSB substitution",
Pattern Recognition, Vol. 37, pp. 469-474 (2004)
[32] Adewole Kayode S, Oladipupo Ayotunde, “Efficient Data Hiding System Using
Cryptography And Steganography”, “International Journal of Applied Information
Systems”, Volume. 4, Dec 2012. ISSN: 2249-0868
[33] J. Silman, “Steganography and steganalysis: an overview,”Tech. Rep., SANS
Institute, 2001.
[34] K. L. Chiew and J. Pieprzyk, “Binary image steganographic techniques
classification based on multi-class steganalysis,” in Information Security, Practice and
Experience. Berlin, Germany: Springer-Verlag, 2010, pp. 341–358.
[35] Yambin Jina Chanu, Themrichon Tuithung, Kh Manglem singh, "A Short Survey
on Image Steganography and Steganalysis Technique," IEEE Trans., 2012
[36] Raftari, N., Moghadam, A. M. E., (2012). Digital Image Steganography Based on
Assignment Algorithm and Combination of DCT-IWT. InComputational Intelligence,
Communication Systems and Networks (CICSyN), 2012 Fourth International Conference
on (pp. 295-300).
[37] Prabakaran, G., Bhavani, R., Kanimozhi, K. (2013). Dual transform based
steganography using wavelet families and statistical methods. In Pattern Recognition,
Informatics and Mobile Engineering (PRIME), 2013 International Conference on (pp.
287-293).
[38] Motamedi, H., Jafari, A. (2012). A new image steganography based on denoising
methods in wavelet domain. In Information Security and Cryptology (ISCISC), 2012 9th
International ISC Conference on (pp. 18-25)..
[39] Ioannidou, A., Halkidis, S. T., Stephanides, G. (2012). A novel technique for
image steganography based on a high payload method and edge detection.Expert Systems
with Applications, 39(14), 11517-11524.
91
[40] Hong, W., Chen, T. S., Chang, Y. P., Shiu, C. W. (2010). A high capacity
reversible data hiding scheme using orthogonal projection and prediction error
modification. Signal Processing, 90(11), 2911-2922.
[41] El-Emam, N., Al-Zubidy, R., (2013). New steganography algorithm to conceal a
large amount of secret message using hybrid adaptive neural networks with modified
adaptive genetic algorithm. Journal of Systems and Software, 86(6), 1465-1481, Elsevier.
[42] Hemalatha, S., Acharya, U. D., Renuka, A., Kamath, P. R. (2013). A secure and
high capacity image steganography technique. Signal & Image Processing, An
International Journal, 4(1), 83-89.
[43] Li, Y. C., Yeh, C. M., Chang, C. C. (2010). Data hiding based on the similarity
between neighbouring pixels with reversibility. Digital Signal Processing, 20(4), 1116-
1128.
[44] Zhu, Y., Yu, M., Hu, H., Ahn, G. J., Zhao, H., (2012). Efficient construction of
provably securesteganography under ordinary covert channels. Science China
Information Sciences, 55(7), 1639-1649.
[45] Wang, X. T., Chang, C. C., Nguyen, T. S., Li, M. C., (2013). Reversible data
hiding for high quality images exploiting interpolation and direction order mechanism.
Digital Signal Processing, 23(2), 569-577.
[46] K. Pramitha, Dr. L.Padma Suresh, K.L.Shunmuganathan, “Image steganography
using mod-4Embedding algorithm based on image Contrast”, Proceedings of 2011
International Conference on Signal Processing, Communication, Computing and
Networking Technologies (ICSCCN 2011)
[47] Sara sajasi, Amir Masoud Eftekhari Moghadam, “A High Quality Image
Steganography Scheme Based on Fuzzy Inference System”, 2013 13th Iranian
Conference on Fuzzy Systems (IFSC)
92
[48] Mansi S. Subhedar, Vijay H. Mankar, “Performance Evaluation of Image
Steganography based on Cover Selection and Contourlet Transform”, 2013 International
Conference on Cloud & Ubiquitous Computing & Emerging Technologies
[49] Mohammad Reza Dastjani Farahani, Ali Pourmohammad, “A DWT Based
Perfect Secure and High Capacity Image Steganography Method”, 2013 International
Conference on Parallel and Distributed Computing, Applications and Technologies
[50] Bingwen Feng, Wei Lu, and Wei Sun, “Secure Binary Image Steganography
Based on Minimizing the Distortion on the Texture”, IEEE TRANSACTIONS ON
INFORMATION FORENSICS AND SECURITY, VOL. 10, NO. 2, FEBRUARY 2015
[51] Vikas Verma, Poonam, Rishma Chawla, “An Enhanced Least Significant Bit
Steganography Method Using Midpoint Circle Approach”, International Conference on
Communication and Signal Processing, April 3-5, 2014, India
[52] Bin Li, Member, IEEE, Ming Wang, Xiaolong Li, Shunquan Tan, Member, IEEE,
and Jiwu Huang, Senior Member, IEEE , “A Strategy of Clustering Modification
Directions in Spatial Image Steganography”, IEEE TRANSACTIONS ON
INFORMATION FORENSICS AND SECURITY, VOL. 10, NO. 9, SEPTEMBER 2015
[53] Saiful Islam, Mangat R Modi and Phalguni Gupta, “Edge-based image
steganography” , Islam et al. EURASIP Journal on Information Security 2014, 2014:8
http://jis.eurasipjournals.com/content/2014/1/8
[54] Y. K. Jain and R. R. Ahirwal, “A Novel Image Steganography Method With
Adaptive Number of Least Significant Bits Modification Based on Private Stego-Keys”,
International Journal of Computer Science and Security (IJCSS), vol. 4, (2010) March 1.
[55] H. Yang, X. Sun and G. Sun, “A High-Capacity Image Data Hiding Scheme
Using Adaptive LSB Substitution”, Journal: Radioengineering, vol. 18, no. 4, (2009), pp.
509-516.
[56] S. Channalli and A. Jadhav, “Steganography an Art of Hiding Data”, International
Journal on Computer Science and Engineering, IJCSE, vol. 1, no. 3, (2009).
[57] C.-H. Yang, C.-Y. Weng, S.-J. Wang, Member, IEEE and H.-M. Sun, “Adaptive
Data Hiding in Edge Areas of Images with Spatial LSB Domain Systems”, IEEE
Transactions on Information Forensics and Security, vol. 3, no. (2008) September 3, pp.
488-497.
93
[58] K.-H. Jung, K.-J. Ha and K.-Y. Yoo, “Image data hiding method based on multi-
pixel differencing and LSB substitution methods”, Proc. 2008 International Conference
on Convergence and Hybrid Information Technology (ICHIT '08), Daejeon (Korea),
(2008) August 28-30, pp. 355-358
[59] H. Zhang, G. Geng and C. Xiong, “Image Steganography Using Pixel-Value
Differencing”, Electronic Commerce and Security, ISECS '09. Second International
Symposium on (2009) May.
[60] W. J. Chen, C. C. Chang and T. H. N. Le, “High Payload Steganography
Mechanism Using Hybrid Edge Detector”, Expert Systems with Applications (ESWA
2010), vol. 37, pp. 3292-3301, (2010) April 4.
[61] V. Madhu Viswanatham and J. Manikonda, “A Novel Technique for Embedding
Data in Spatial Domain”, International Journal on Computer Science and Engineering,
IJCSE, vol. 2, (2010).
[62] M. A. Al-Husainy, “Image Steganography by Mapping Pixels to Letters”, Journal
of Computer Science, vol. 5, no. 1, (2009), pp. 33-38.
[63] H. Motameni, M. Norouzi, M. Jahandar and A. Hatami, “Labeling Method in
Steganography”, World Academy of Science, Engineering and Technology, France,
(2007).
[64] M. Tanvir Parvez and A. Abdul-Aziz Gutub, “RGB Intensity Based Variable-Bits
Image Steganography”, IEEE Asia-Pacific Services Computing Conference, (2008), pp.
1322-1327.
[65] A. M. Hamid and M. L. M. Kiah, “Novel Approach for High Secure and High
Rate Data Hidden in the Image Using Image Texture Analysis”, International Journal of
Engineering and Technology (IJET): 0975-4042, (2009).
[66] M. Chaumont and W. Puech, “DCT-Based Data Hiding Method To Embed the
Color Information in a JPEG Grey Level Image”, 14th European Signal Processing
Conference (EUSIPCO 2006), Florence, Italy, copyright by EURASIP, (2006) September
4-8.
[67] K. S. Babu, K. B. Raja, K. Kiran Kumar, T. H. Manjula Devi, K. R. Venugopal
and L. M. Pataki, “Authentication of secret information in image steganography”, IEEE
Region 10 Conference, TENCON-2008, (2008) November, pp. 1-6.
94
[68] Mamta Juneja, Parvinder Singh Sandhu, “Designing of Robust Image
Steganography Technique Based on LSB Insertion and Encryption”, 2009 International
Conference on Advances in Recent Technologies in Communication and Computing
[69] Subba Rao Y.V, Brahmananda Rao S.S, Rukma Rekha N , “Secure Image
Steganography based on Randomized Sequence of Cipher Bits”, 2011 Eighth
International Conference on Information Technology: New Generations
[70] Septimiu Fabian Mare, Mircea Vladutiu and Lucian Prodan, “Secret data
communication system using Steganography, AES and RSA”, 2011 IEEE 17th
International Symposium for Design and Technology in Electronic Packaging (SIITME)
[71] S Ushll, G A Sathish Kumal, K Boopathybagan, “A Secure Triple Level
Encryption Method Using Cryptography and Steganography”, 2011 International
Conference on Computer Science and Network Technology
[72] Yang Ren-er, Zheng Zhiwei, Tao Shun, Ding Shilei, “Image Steganography
Combined with DES Encryption Pre-processing”, 2014 Sixth International Conference on
Measuring Technology and Mechatronics Automation
[73] Diaa E.M.Ahmed, Othman O.Khalifa, “ Robust and Secure Image Steganography
Based on Elliptic Curve Cryptography”, 5th International Conference on Computer &
Communication Engineering
[74] Dipanwita Debnath, Suman Deb, Nirmalya Kar, “An Advanced Image Encryption
Standard Providing Dual Security: encryption using Hill Cipher & RGB image
steganography”, 2015 International Conference on Computational Intelligence &
Networks
[75] Jeng-Shyang Pan, Wei Li, Chun-Sheng Yang , Li-Jun Yan, “Image
steganography based on subsampling and compressive sensing, Springer Science
Business Media New York 2014
[76] Nitin Kaul, Nikesh Bajaj, “,”Audio in Image Steganography based on Wavelet
Transform”, International Journal of Computer Applications (0975 – 8887) Volume 79 –
No3, October 2013
[77] Amol Bhujade, Prof. Sonu Lal, “Advanced Steganography: Embedding
HighCapacity Audio in Colour Image”, International Journal of Advanced Research in
95
Electrical, Electronics and Instrumentation Engineering (An ISO 3297: 2007 Certified
Organization) Vol . 4, Issue 7, July 2015
[78] Devendra Singh Rao, Pankaj Singh Parihar, “Embedding Approach of Audio
Data in RGB Images Using Circle Equation”, International Journal on Recent and
Innovation Trends in Computing and Communication, Volume:2 , Issue 7
[79] RAMESH GOTTIPATI, “Audio Based Security System with Image
Steganography”, International Journal of Software Engineering and Technology
Informatics Volume. 01, IssueNo.01, January-2015, Pages: 0001-0005
[80] E. J. Candès, J. Romberg, and T. Tao, ``Robust uncertainty principles: Exact
signal reconstruction from highly incomplete frequency information,'' IEEE Trans. Inf.
Theory, vol. 52, no. 2, pp. 489_509, Feb. 2006.
[81] D. L. Donoho, ``Compressed sensing,'' IEEE Trans. Inf. Theory, vol. 52, no. 4,
pp. 1289_1306, Apr. 2006.
[82] E. J. Candès and M. B. Wakin, ``An introduction to compressive sampling,''
IEEE Signal Process. Mag., vol. 25, no. 2, pp. 21_30,Mar. 2008.
[83] G. Wunder, H. Boche, T. Strohmer, and P. Jung, ``Sparse signal processing
concepts for ef_cient 5G system design,'' IEEE Access, vol. 3, pp. 195_208, 2015.
[84] H. Kung, C.-K. Lin, and D. Vlah, ``CloudSense: Continuous _ne-grain cloud
monitoring with compressive sensing,'' in Proc. HotCloud, 2011, pp. 1_6.
[85] X. Liu, D. Zhai, J. Zhou, X. Zhang, D. Zhao, andW. Gao, ``Compressive
sampling-based image coding for resource-de_cient visual communication,'' IEEE Trans.
Image Process., vol. 25, no. 6, pp. 2844_2855, Jun. 2016.
[86] E. J. Candès and T. Tao, ``Near-optimal signal recovery from random
projections: Universal encoding strategies?'' IEEE Trans. Inf. Theory, vol. 52, no. 12, pp.
5406_5425, Dec. 2006.
[87] Y. Rachlin and D. Baron, ``The secrecy of compressed sensing measurements,'' in
Proc. 46th Annu. Allerton Conf. Commun., Control, Comput., Urbana, IL, USA, Sep.
2008, pp. 813_817.
[88] A. Orsdemir, H. O. Altun, G. Sharma, and M. F. Bocko, ``On the security and
robustness of encryption via compressed sensing,'' in Proc. IEEE Military Commun.
Conf. (MILCOM), San Diego, CA, USA, Nov. 2008, pp. 1_7.
96
[89] S. A. Hossein, A. E. Tabatabaei, and N. Zivic, ``Security analysis of the joint
encryption and compressed sensing,'' in Proc. 20th Telecommun. Forum (TELFOR),
Belgrade, Serbia, Nov. 2012, pp. 799_802.
[90] T. Bianchi, V. Bioglio, and E. Magli, ``On the security of random linear
measurements,'' in Proc. IEEE Int. Conf. Acoust., Speech Signal Process. (ICASSP),
Florence, Italy, May 2014, pp. 3992_3996.
[91] T. Bianchi, V. Bioglio, and E. Magli, ``Analysis of one-time random projections
for privacy preserving compressed sensing,'' IEEE Trans. Inf. Forensics Security, vol. 11,
no. 2, pp. 313_327, Feb. 2016.
[92] S. Agrawal and S. Vishwanath, ``Secrecy using compressive sensing,'' in Proc.
IEEE Inf. Theory Workshop (ITW), Paraty, Brazil, Oct. 2011, pp. 563_567.
[93] Simon Forcart and Holger Rauhut, “A Mathematical Introduction to Compressive
Sensing”.
[94] Gitta Kutyniok, “Theory and Applications of Compressed Sensing”, GAMM-
Mitteilungen, 10 July 2013
[95] E. J. Candès and M. B. Wakin, ``An introduction to compressive sampling,'' IEEE
Signal Process. Mag., vol. 25, no. 2, pp. 21_30,
Mar. 2008.
[96] G. Wunder, H. Boche, T. Strohmer, and P. Jung, ``Sparse signal processing
concepts for ef_cient 5G system design,'' IEEE Access, vol. 3, pp. 195_208, 2015.
[97] H. Kung, C.-K. Lin, and D. Vlah, ``CloudSense: Continuous _ne-grain cloud
monitoring with compressive sensing,'' in Proc. HotCloud, 2011, pp. 1_6.
X. Liu, D. Zhai, J. Zhou, X. Zhang, D. Zhao, andW. Gao, ``Compressive sampling-based
image coding for resource-de_cient visual communication,'' IEEE Trans. Image Process.,
vol. 25, no. 6, pp. 2844_2855, Jun. 2016
[98] Wafaa A. Shalaby, Waleed Saad, Mona Shokair, Moawad Dessouky, “Efficient
Parameters for Compressed Sensing Recovery Algorithms”, Wireless Pers Commun
(2017) 94:1715–1736
[99] . Hayashi, K., Nagahara, N., & Tanaka, T. (2013). “A user’s guide to compressed
sensing for communication system”. IEICE Transactions on Communications, 96(3),
685–712.
97
[100] Gribonval, R., & Nielsen, M., “ Sparse representations in unions of bases” IEEE
Transactions on Information Theory, 49(12), 3320–3325.(2003)
[101] Blvmensath, T., & Davies, M. “Iterative hard thresholding for compressed
sensing. Applied and Computational Harmonic Analysis”, 27(3), 265–274.(2009)
[102] D. L. Donoho, ``Compressed sensing,'' IEEE Trans. Inf. Theory, vol. 52, no. 4, pp.
1289_1306, Apr. 2006.
[103] Daemen J, Rijmen V. “The Design of RIJNDAEL: AES —the Advanced
Encryption Standard”. Springer- Verlag: Berlin, 2002.
[104] Murphy S, Robshaw MJ. “Essential algebraic structure within the AES”.
Proceedings of the 22th Annual International Cryptology, Springer-Verlag, 2002; 1–16.
[105] Liu J, Wai B, Cheng X, Wang X. “An AES S-box to increase complexity and
cryptographic analysis”. Proceedings of the 19th International Conference on Advanced
Information Networking and Applications, Taiwan, 2005; 724–728.
[106] Yang Ren-er, Zheng Zhiwei, Tao Shun, Ding Shilei, “ Image Steganography
Combined with DES Encryption Standard”. 6th International Conference on Measuring
Technology and Mechatronics Automation 2014
[107] Kocarev L., Jakimoski G., Stojanovski T., and Parlitz U., “From Chaotic Maps to
Encryption Schemes,” in Proceedings of IEEE International Symposium Circuits and
Systems, pp. 514-517, 1998.
[108] Behnia S., Akhshani A., Ahadpour S., and Mahmodi H., “A Fast Chaotic
Encryption Scheme Based on Piecewise Nonlinear Chaotic Maps,” Physics Letters A,
vol. 366, no. 4-5, pp. 391-396, 2007.
[109] Schneier B., Applied Cryptography: Protocols, Algorithms, and Source Code in
C, John Wiley and Sons, New York, 1996.
[110] E. Biham, “Cryptanalysis of the chaotic-map cryptosystem,” Proc. Eurocrypt ’91,
pp. 532-534, 1991.
[111] Shai Halevi and Hugo Krawczyk, “Public-key cryptography and password
protocols,” ACM Transactions on Information and System Security, pp. 230-268, August
1999.
[112] M. R. K. Ariffin and N. A. Abu, “A Chaos Based Public Key Cryptosystem,”
International Journal of Cryptology Research, pp.149-163, 2009.
98
[113] L. Kocarev, S.Lian (Eds.), “Chaos-Based Cryptography - Theory, Algorithms and
Applications”, Studies in computational Intelligence vol. 354, Springer, 2011.
[114] L. Kocarev and M. Sterjev, “Public key encryption scheme with chaos,” Chaos,
vol. 14, pp.1078-1081, 2004.
[115] Y. Zhang, F. Zuo, Z. Zhai, and C. Xiaobin, “A new image encryption algorithm
based on multiple chaos system,” in Proceedings of the International Symposium on
Electronic Commerce and Security (ISECS ’08), pp. 347–350, August 2008.
[116] Mubashar Khan, Naveed Ahmed Azam,” Right translated AES gray S-boxes”.
Security Comm. Networks 2015
[117] Lifang Yu, Yao Zhao, Rongrong Ni (EURAISP Member), and Ting Li,”
Improved Adaptive LSB Stegnography Based on Chaos and Genetic
Algorithm”.EURASIP Journal on Advances in Signal Processing Volume:2010
[118] James Collins, Sos Agaian, ” High Capacity Image Steganography Using
Adjunctive Numerical Representations With Multiple Bit-Plane Decomposition
Methods” International Journal on Cryptography and Information Security (IJCIS), Vol.
6, No. 1/2, June 2016
[119] Anita Pradhan, Aditya Kumar Sahu, Gandharba Swain, K. Raja Sekhar, “
Performance Evaluation Parameters of Image Steganography Techniques, International
Conference on Research Advances in Integrated Navigation Systems (RAINS - 2016),
April 06-07, 2016, R. L. Jalappa Institute of Technology, Doddaballapur, Bangalore,
India
[120] Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. "Image quality
assessment: from error visibility to structural similarity." Image Processing, IEEE
Transactions on 13.4 (2004): 600-612.